Referensi
Bidang profil
Referensi yang dihasilkan dari sumber untuk setiap bidang dalam skema profil bersama Apothem.
Setiap bidang di bawah ini dihasilkan dari JSON Schema profil kanonis di
src/apothem/schemas/profile.schema.json. Tabel ini adalah sumber kebenaran
untuk apa yang boleh dideklarasikan sebuah profil bersama; sunting skema, lalu hasilkan ulang.
Referensi Bidang Profil yang Dihasilkan
| Field | Type | Required | Description |
|---|---|---|---|
agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
exclude_harnesses | array | no | Supported public harness ids skipped by batch lifecycle commands. |
harnesses | object | no | Per-harness overrides keyed by supported public harness id. |
harnesses.antigravity | object | no | Canonical profile fields overridden for one harness. |
harnesses.antigravity.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.antigravity.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.antigravity.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.antigravity.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.antigravity.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.antigravity.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.antigravity.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.antigravity.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.antigravity.identity.email | string | no | — |
harnesses.antigravity.identity.github | string | no | — |
harnesses.antigravity.identity.name | string | no | — |
harnesses.antigravity.identity.role | string | no | — |
harnesses.antigravity.identity.website | string | no | — |
harnesses.antigravity.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.antigravity.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.antigravity.preferences.formatter | string | no | Preferred code formatter. |
harnesses.antigravity.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.antigravity.preferences.style | string | no | Preferred AI response style. |
harnesses.antigravity.preferences.test_framework | string | no | Preferred test framework. |
harnesses.antigravity.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.antigravity.seriousness | string | no | Governance seriousness tier. |
harnesses.claude-code | object | no | Canonical profile fields overridden for one harness. |
harnesses.claude-code.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.claude-code.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.claude-code.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.claude-code.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.claude-code.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.claude-code.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.claude-code.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.claude-code.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.claude-code.identity.email | string | no | — |
harnesses.claude-code.identity.github | string | no | — |
harnesses.claude-code.identity.name | string | no | — |
harnesses.claude-code.identity.role | string | no | — |
harnesses.claude-code.identity.website | string | no | — |
harnesses.claude-code.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.claude-code.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.claude-code.preferences.formatter | string | no | Preferred code formatter. |
harnesses.claude-code.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.claude-code.preferences.style | string | no | Preferred AI response style. |
harnesses.claude-code.preferences.test_framework | string | no | Preferred test framework. |
harnesses.claude-code.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.claude-code.seriousness | string | no | Governance seriousness tier. |
harnesses.codebuddy | object | no | Canonical profile fields overridden for one harness. |
harnesses.codebuddy.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.codebuddy.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.codebuddy.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.codebuddy.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.codebuddy.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.codebuddy.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.codebuddy.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.codebuddy.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.codebuddy.identity.email | string | no | — |
harnesses.codebuddy.identity.github | string | no | — |
harnesses.codebuddy.identity.name | string | no | — |
harnesses.codebuddy.identity.role | string | no | — |
harnesses.codebuddy.identity.website | string | no | — |
harnesses.codebuddy.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.codebuddy.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.codebuddy.preferences.formatter | string | no | Preferred code formatter. |
harnesses.codebuddy.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.codebuddy.preferences.style | string | no | Preferred AI response style. |
harnesses.codebuddy.preferences.test_framework | string | no | Preferred test framework. |
harnesses.codebuddy.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.codebuddy.seriousness | string | no | Governance seriousness tier. |
harnesses.codex | object | no | Canonical profile fields overridden for one harness. |
harnesses.codex.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.codex.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.codex.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.codex.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.codex.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.codex.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.codex.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.codex.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.codex.identity.email | string | no | — |
harnesses.codex.identity.github | string | no | — |
harnesses.codex.identity.name | string | no | — |
harnesses.codex.identity.role | string | no | — |
harnesses.codex.identity.website | string | no | — |
harnesses.codex.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.codex.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.codex.preferences.formatter | string | no | Preferred code formatter. |
harnesses.codex.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.codex.preferences.style | string | no | Preferred AI response style. |
harnesses.codex.preferences.test_framework | string | no | Preferred test framework. |
harnesses.codex.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.codex.seriousness | string | no | Governance seriousness tier. |
harnesses.cursor | object | no | Canonical profile fields overridden for one harness. |
harnesses.cursor.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.cursor.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.cursor.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.cursor.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.cursor.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.cursor.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.cursor.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.cursor.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.cursor.identity.email | string | no | — |
harnesses.cursor.identity.github | string | no | — |
harnesses.cursor.identity.name | string | no | — |
harnesses.cursor.identity.role | string | no | — |
harnesses.cursor.identity.website | string | no | — |
harnesses.cursor.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.cursor.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.cursor.preferences.formatter | string | no | Preferred code formatter. |
harnesses.cursor.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.cursor.preferences.style | string | no | Preferred AI response style. |
harnesses.cursor.preferences.test_framework | string | no | Preferred test framework. |
harnesses.cursor.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.cursor.seriousness | string | no | Governance seriousness tier. |
harnesses.gemini-cli | object | no | Canonical profile fields overridden for one harness. |
harnesses.gemini-cli.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.gemini-cli.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.gemini-cli.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.gemini-cli.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.gemini-cli.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.gemini-cli.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.gemini-cli.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.gemini-cli.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.gemini-cli.identity.email | string | no | — |
harnesses.gemini-cli.identity.github | string | no | — |
harnesses.gemini-cli.identity.name | string | no | — |
harnesses.gemini-cli.identity.role | string | no | — |
harnesses.gemini-cli.identity.website | string | no | — |
harnesses.gemini-cli.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.gemini-cli.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.gemini-cli.preferences.formatter | string | no | Preferred code formatter. |
harnesses.gemini-cli.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.gemini-cli.preferences.style | string | no | Preferred AI response style. |
harnesses.gemini-cli.preferences.test_framework | string | no | Preferred test framework. |
harnesses.gemini-cli.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.gemini-cli.seriousness | string | no | Governance seriousness tier. |
harnesses.github-copilot | object | no | Canonical profile fields overridden for one harness. |
harnesses.github-copilot.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.github-copilot.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.github-copilot.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.github-copilot.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.github-copilot.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.github-copilot.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.github-copilot.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.github-copilot.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.github-copilot.identity.email | string | no | — |
harnesses.github-copilot.identity.github | string | no | — |
harnesses.github-copilot.identity.name | string | no | — |
harnesses.github-copilot.identity.role | string | no | — |
harnesses.github-copilot.identity.website | string | no | — |
harnesses.github-copilot.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.github-copilot.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.github-copilot.preferences.formatter | string | no | Preferred code formatter. |
harnesses.github-copilot.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.github-copilot.preferences.style | string | no | Preferred AI response style. |
harnesses.github-copilot.preferences.test_framework | string | no | Preferred test framework. |
harnesses.github-copilot.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.github-copilot.seriousness | string | no | Governance seriousness tier. |
harnesses.glm | object | no | Canonical profile fields overridden for one harness. |
harnesses.glm.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.glm.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.glm.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.glm.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.glm.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.glm.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.glm.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.glm.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.glm.identity.email | string | no | — |
harnesses.glm.identity.github | string | no | — |
harnesses.glm.identity.name | string | no | — |
harnesses.glm.identity.role | string | no | — |
harnesses.glm.identity.website | string | no | — |
harnesses.glm.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.glm.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.glm.preferences.formatter | string | no | Preferred code formatter. |
harnesses.glm.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.glm.preferences.style | string | no | Preferred AI response style. |
harnesses.glm.preferences.test_framework | string | no | Preferred test framework. |
harnesses.glm.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.glm.seriousness | string | no | Governance seriousness tier. |
harnesses.hermes | object | no | Canonical profile fields overridden for one harness. |
harnesses.hermes.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.hermes.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.hermes.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.hermes.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.hermes.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.hermes.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.hermes.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.hermes.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.hermes.identity.email | string | no | — |
harnesses.hermes.identity.github | string | no | — |
harnesses.hermes.identity.name | string | no | — |
harnesses.hermes.identity.role | string | no | — |
harnesses.hermes.identity.website | string | no | — |
harnesses.hermes.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.hermes.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.hermes.preferences.formatter | string | no | Preferred code formatter. |
harnesses.hermes.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.hermes.preferences.style | string | no | Preferred AI response style. |
harnesses.hermes.preferences.test_framework | string | no | Preferred test framework. |
harnesses.hermes.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.hermes.seriousness | string | no | Governance seriousness tier. |
harnesses.kimi-code | object | no | Canonical profile fields overridden for one harness. |
harnesses.kimi-code.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.kimi-code.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.kimi-code.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.kimi-code.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.kimi-code.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.kimi-code.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.kimi-code.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.kimi-code.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.kimi-code.identity.email | string | no | — |
harnesses.kimi-code.identity.github | string | no | — |
harnesses.kimi-code.identity.name | string | no | — |
harnesses.kimi-code.identity.role | string | no | — |
harnesses.kimi-code.identity.website | string | no | — |
harnesses.kimi-code.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.kimi-code.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.kimi-code.preferences.formatter | string | no | Preferred code formatter. |
harnesses.kimi-code.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.kimi-code.preferences.style | string | no | Preferred AI response style. |
harnesses.kimi-code.preferences.test_framework | string | no | Preferred test framework. |
harnesses.kimi-code.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.kimi-code.seriousness | string | no | Governance seriousness tier. |
harnesses.kiro | object | no | Canonical profile fields overridden for one harness. |
harnesses.kiro.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.kiro.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.kiro.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.kiro.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.kiro.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.kiro.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.kiro.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.kiro.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.kiro.identity.email | string | no | — |
harnesses.kiro.identity.github | string | no | — |
harnesses.kiro.identity.name | string | no | — |
harnesses.kiro.identity.role | string | no | — |
harnesses.kiro.identity.website | string | no | — |
harnesses.kiro.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.kiro.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.kiro.preferences.formatter | string | no | Preferred code formatter. |
harnesses.kiro.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.kiro.preferences.style | string | no | Preferred AI response style. |
harnesses.kiro.preferences.test_framework | string | no | Preferred test framework. |
harnesses.kiro.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.kiro.seriousness | string | no | Governance seriousness tier. |
harnesses.open-claw | object | no | Canonical profile fields overridden for one harness. |
harnesses.open-claw.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.open-claw.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.open-claw.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.open-claw.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.open-claw.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.open-claw.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.open-claw.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.open-claw.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.open-claw.identity.email | string | no | — |
harnesses.open-claw.identity.github | string | no | — |
harnesses.open-claw.identity.name | string | no | — |
harnesses.open-claw.identity.role | string | no | — |
harnesses.open-claw.identity.website | string | no | — |
harnesses.open-claw.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.open-claw.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.open-claw.preferences.formatter | string | no | Preferred code formatter. |
harnesses.open-claw.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.open-claw.preferences.style | string | no | Preferred AI response style. |
harnesses.open-claw.preferences.test_framework | string | no | Preferred test framework. |
harnesses.open-claw.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.open-claw.seriousness | string | no | Governance seriousness tier. |
harnesses.opencode | object | no | Canonical profile fields overridden for one harness. |
harnesses.opencode.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.opencode.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.opencode.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.opencode.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.opencode.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.opencode.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.opencode.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.opencode.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.opencode.identity.email | string | no | — |
harnesses.opencode.identity.github | string | no | — |
harnesses.opencode.identity.name | string | no | — |
harnesses.opencode.identity.role | string | no | — |
harnesses.opencode.identity.website | string | no | — |
harnesses.opencode.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.opencode.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.opencode.preferences.formatter | string | no | Preferred code formatter. |
harnesses.opencode.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.opencode.preferences.style | string | no | Preferred AI response style. |
harnesses.opencode.preferences.test_framework | string | no | Preferred test framework. |
harnesses.opencode.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.opencode.seriousness | string | no | Governance seriousness tier. |
harnesses.qwen-code | object | no | Canonical profile fields overridden for one harness. |
harnesses.qwen-code.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.qwen-code.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.qwen-code.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.qwen-code.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.qwen-code.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.qwen-code.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.qwen-code.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.qwen-code.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.qwen-code.identity.email | string | no | — |
harnesses.qwen-code.identity.github | string | no | — |
harnesses.qwen-code.identity.name | string | no | — |
harnesses.qwen-code.identity.role | string | no | — |
harnesses.qwen-code.identity.website | string | no | — |
harnesses.qwen-code.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.qwen-code.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.qwen-code.preferences.formatter | string | no | Preferred code formatter. |
harnesses.qwen-code.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.qwen-code.preferences.style | string | no | Preferred AI response style. |
harnesses.qwen-code.preferences.test_framework | string | no | Preferred test framework. |
harnesses.qwen-code.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.qwen-code.seriousness | string | no | Governance seriousness tier. |
harnesses.trae | object | no | Canonical profile fields overridden for one harness. |
harnesses.trae.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.trae.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.trae.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.trae.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.trae.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.trae.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.trae.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.trae.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.trae.identity.email | string | no | — |
harnesses.trae.identity.github | string | no | — |
harnesses.trae.identity.name | string | no | — |
harnesses.trae.identity.role | string | no | — |
harnesses.trae.identity.website | string | no | — |
harnesses.trae.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.trae.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.trae.preferences.formatter | string | no | Preferred code formatter. |
harnesses.trae.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.trae.preferences.style | string | no | Preferred AI response style. |
harnesses.trae.preferences.test_framework | string | no | Preferred test framework. |
harnesses.trae.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.trae.seriousness | string | no | Governance seriousness tier. |
harnesses.windsurf | object | no | Canonical profile fields overridden for one harness. |
harnesses.windsurf.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.windsurf.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.windsurf.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.windsurf.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.windsurf.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.windsurf.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.windsurf.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.windsurf.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.windsurf.identity.email | string | no | — |
harnesses.windsurf.identity.github | string | no | — |
harnesses.windsurf.identity.name | string | no | — |
harnesses.windsurf.identity.role | string | no | — |
harnesses.windsurf.identity.website | string | no | — |
harnesses.windsurf.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.windsurf.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.windsurf.preferences.formatter | string | no | Preferred code formatter. |
harnesses.windsurf.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.windsurf.preferences.style | string | no | Preferred AI response style. |
harnesses.windsurf.preferences.test_framework | string | no | Preferred test framework. |
harnesses.windsurf.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.windsurf.seriousness | string | no | Governance seriousness tier. |
harnesses.zed | object | no | Canonical profile fields overridden for one harness. |
harnesses.zed.agent_mcp_servers | object | no | Per subagent, the MCP servers that agent may call, keyed by agent name (the file stem under agents/, for example citation-auditor). Safe by default: with no entry an agent is granted no MCP tool. The Claude Code engine install adds one wildcard entry per server (that server's tools, and no other server's) to the agent's explicit tools list, and only for an agent whose source has one. Write each server name as /mcp shows it. Claude Code writes a colon or period as an underscore in tool names, so the install renders plugin:bio-research:biorxiv as mcp__plugin_bio-research_biorxiv__. A name allows letters, digits, hyphen, underscore, colon and period, begins and ends with a letter or digit, holds no two of underscore, colon and period in a row, no asterisk and no whitespace, and is at most 64 characters. A per-harness override unions with the shared lists. The Gemini CLI install adds one mcp_<server>_ entry per server to the explicit tools list of the listed agents, for a server name of letters, digits and hyphen (Gemini CLI cannot address a name with an underscore, so such a name is skipped and reported). Harnesses other than Claude Code and Gemini CLI do not read this key. |
harnesses.zed.enforcement | object | no | Opt-in toggles for shipped behaviors that ship default-off; every flag defaults false and a clean install imposes none of them. |
harnesses.zed.enforcement.agent_teams | boolean | no | Opt in to parallel agent-team dispatch for parallelizable work. |
harnesses.zed.enforcement.continuous_execution | boolean | no | Opt in to continuous multi-step advancement across natural boundaries. |
harnesses.zed.enforcement.learning_loop | boolean | no | Opt in to continuous-learning capture and pattern promotion. |
harnesses.zed.enforcement.multitasking | boolean | no | Opt in to parallel multi-task execution. |
harnesses.zed.enforcement.sprints | boolean | no | Opt in to the sprint apparatus for non-trivial multi-step work. |
harnesses.zed.identity | object | no | Optional identity fields overridden for one harness. |
harnesses.zed.identity.email | string | no | — |
harnesses.zed.identity.github | string | no | — |
harnesses.zed.identity.name | string | no | — |
harnesses.zed.identity.role | string | no | — |
harnesses.zed.identity.website | string | no | — |
harnesses.zed.mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
harnesses.zed.preferences | object | no | Workflow and interaction preferences shared across harnesses. |
harnesses.zed.preferences.formatter | string | no | Preferred code formatter. |
harnesses.zed.preferences.language | string | no | Primary programming language in lower-case semantic form. |
harnesses.zed.preferences.style | string | no | Preferred AI response style. |
harnesses.zed.preferences.test_framework | string | no | Preferred test framework. |
harnesses.zed.rules | array | no | Additional rules shared with capable harnesses. |
harnesses.zed.seriousness | string | no | Governance seriousness tier. |
identity | object | yes | Operator identity fields shared across harnesses. |
identity.email | string | no | Operator email address. |
identity.github | string | no | GitHub username without @. |
identity.name | string | yes | Operator display name. |
identity.role | string | no | Operator role presented to capable harnesses. |
identity.website | string | no | Operator personal or project website URL. |
mcp_servers | object | no | MCP server inventory keyed by server name: the canonical cross-harness portability surface for Model Context Protocol servers. |
preferences | object | no | Workflow and interaction preferences shared across harnesses. |
preferences.formatter | string | no | Preferred code formatter. |
preferences.language | string | no | Primary programming language in lower-case semantic form. |
preferences.style | string | no | Preferred AI response style. |
preferences.test_framework | string | no | Preferred test framework. |
rules | array | no | Additional rules shared with capable harnesses. |
schema_version | integer | no | Optional profile-schema version. A profile without it is read as version 1 and migrated forward like any v1 profile; profile scaffolds always write it. A profile stamped with a version greater than the engine supports is rejected with an upgrade-the-engine diagnostic before validation. |
seriousness | string | no | Governance seriousness tier. |
workspace | object | no | Where Apothem keeps its shared, gitignored working directory (plans, memory, learning, contexts). One directory shared across every harness; never per-harness. |
workspace.directory_name | string | no | Name of the shared working directory. A single path component (no separator, no drive specifier). Defaults to '.apothem'. |
workspace.scope | string | no | Where the working directory is rooted. 'project-local' places <project-root>/<directory_name>/ (default); 'user-home' places ~/<directory_name>/. |