Référence
Règles
Règles comportementales toujours actives et filtrées par chemin régissant la résolution de problèmes, l'artisanat du code, l'architecture propre, la planification et la qualité à travers l'ensemble des règles comportementales.
Les règles comportementales vivent sous forme de fichiers plats sous
src/apothem/rules/, avec le suffixe *.md. Chaque règle est autonome —
Purpose, Obligations, Seriousness Scaling, Anti-Patterns, Enforcement,
Bindings — et est soit toujours active, soit filtrée par chemin.
Règles actives
Le registre canonique est la cohorte de règles dans src/apothem/rules/.
L'inventaire généré au bas de cette page liste
chaque règle avec sa description et son fichier source. Cinq règles sont
toujours actives et les autres sont filtrées par chemin ; la section suivante
nomme les cinq qui se chargent à chaque session.
Contrat de frontmatter
Chaque fichier de règle porte :
---
name: <rule name>
description: <one-line role>
pathFilter: <glob list; empty string "" on always-on rules>
alwaysApply: <true | false>
---Sur une règle toujours active, pathFilter est la chaîne vide ("") et
alwaysApply vaut true ; sur une règle filtrée par chemin, pathFilter porte
la liste de globs et alwaysApply vaut false. Le contrat lisible par machine
est
frontmatter_grep
({name, description, pathFilter, alwaysApply}) ; la vérification
frontmatter-cohort du validateur le vérifie sur chaque fichier de règle.
pathFilter et alwaysApply sont neutres vis-à-vis de l'environnement. Chaque
adaptateur les traduit là où son environnement dispose d'un champ de portée
natif. Claude Code ne lit que paths: dans une règle et charge au lancement
toute règle qui en est dépourvue, comme il le fait pour une règle dont paths:
vaut ** seul ; l'installation par le moteur
(apothem install --harness claude-code) écrit donc chaque règle filtrée par
chemin dans ~/.claude/rules/ avec une liste paths: construite à partir de
son pathFilter, et écrit chaque copie de règle sans sa section Bindings
finale. Cinq règles se chargent à chaque session : agnostic-posture,
authority-inquiry, session-closure, context-management et l'index des
règles. Les 90 autres règles sont filtrées par chemin et se chargent lorsque
l'assistant lit un fichier correspondant. Ajouter une règle à l'ensemble
toujours actif modifie la liste de base et le plafond d'octets dans le même
changement.
L'index des règles, apothem-rule-index, est l'unique fichier toujours actif
qui nomme chaque autre règle et la situation qui l'appelle. Un environnement
qui délimite les règles par chemin charge une règle délimitée lorsqu'un fichier
correspondant est touché. Un environnement qui ne le fait pas n'en charge
aucune, de sorte que l'index, la phrase du fichier d'instructions qui pointe
vers lui, et les règles et skills qui citent une règle par son chemin sont la
façon dont une session l'atteint. Après une compaction du contexte, une règle
délimitée n'est plus dans le contexte, et l'index indique à l'assistant de
relire toute règle qu'il appliquait. Claude Code ne compare un filtre de chemin
qu'aux fichiers situés dans le répertoire de travail, de sorte que les règles
de mémoire automatique, dont les fichiers vivent sous ~/.claude/projects/,
atteignent une session par l'index et non par une correspondance de chemin.
OpenCode ne dispose d'aucun champ de portée par règle : il combine dans chaque
session tous les fichiers que nomme la liste instructions de son
opencode.json. L'installation OpenCode (apothem install --harness opencode)
liste donc votre document de profil et uniquement les règles dont alwaysApply
vaut true (les quatre règles de base et l'index des règles), à raison d'un
chemin ~/.config/opencode/... explicite par fichier, lu depuis les fichiers
de règles au moment où la configuration est écrite. Chaque règle, y compris les
règles filtrées par chemin, est tout de même installée sous
~/.config/opencode/.apothem/support/rules/, où l'index la nomme et où une
règle, une skill ou une commande qui en cite une la lit à la demande.
Discipline de rédaction
- Autonome. Chaque règle possède ses propres sections Purpose, Obligations, Anti-Patterns et Enforcement.
- Bloc Bindings. Chaque règle se termine par le bloc Bindings canonique à cinq directions (Drives → / Satisfies → / Established by ↑ / Gated by ← / Cross-bound with ↔).
- Langage de domaine uniquement. Aucune référence interne au plan — voir
CLAUDE.mdCM-7.
Inventaire généré des règles
| Name | Description | Source |
|---|---|---|
README | Behavioral instruction rules — flat .md files, each a self-contained directive set the harness loads to govern agent behavior. Each carries YAML frontmatter declaring name, `de | src/apothem/rules/README.md |
agent-capability-discipline-matrix | Path-filtered companion to agent-capability-discipline.md — carries the per-harness agentic-capability matrix (§1), the per-harness MCP-surface catalog (§3), and the per-harness agent-memory-convention catalog (§7) the parent rule's anchors delegate to. Demand-loaded when the assistant edits any adapter sub-package, cross-harness capability-matrix input, or per-harness MCP-config artifact. | src/apothem/rules/agent-capability-discipline-matrix.md |
agent-capability-discipline | Cross-harness agentic discipline — the 17-harness adapter cohort is recognized as sophisticated AI agent systems; core and supplemental agentic capabilities converge via M14, anchor to per-adapter STANDARD CONVENTION PIN, validate at materializer time, and attest through adapter capability coverage. Demand-loaded on adapter / capability / MCP-config touches, co-triggering with its matrix companion. | src/apothem/rules/agent-capability-discipline.md |
agent-orchestration-patterns | Path-filtered companion sub-rule carrying the team-pattern catalog, agent-type selection, launch protocol, return-contract enforcement, isolation discipline, error handling, decision tree, and anti-patterns that the parent agent-orchestration.md rule's anchors declare. | src/apothem/rules/agent-orchestration-patterns.md |
agent-orchestration | Agent and agent team orchestration — the six canonical team patterns (Research, Audit, Implementation, Generation, Quality, Documentation), the deploy-when threshold (3+ independent parallel operations, multi-path exploration, heavy reads, multi-dimension verification, multi-file generation), the single-message parallel-launch invariant, explicit return contracts, and context isolation. Canonical specification for CM-17 / CM-25. | src/apothem/rules/agent-orchestration.md |
agents-md-convention | The repository carries a single agent-facing canon at the root AGENTS.md; per-folder operating guidance lives in each folder's README.md, which serves both the human and the agent reader. Per-folder AGENTS.md companions are not required; any present companion stays current with its folder and the root AGENTS.md stays coherent with the AI-surface canon. | src/apothem/rules/agents-md-convention.md |
agile-sprints-elements | Path-filtered companion rule carrying the canonical sprint-element bodies (Sprint Goal, Backlog INVEST, DoR, DoD, Review, Retrospective, Velocity), the Three Pillars detail, the Surface Imposition table, and the Failure Recovery table; demand-loaded when the parent agile-sprints.md rule's anchors surface. | src/apothem/rules/agile-sprints-elements.md |
agile-sprints | Non-trivial multi-step work in host projects runs as disciplined Agile sprints — Sprint Goal + Backlog + Definition of Ready + Definition of Done + Sprint Review + Retrospective + Velocity tracking. The trivial-vs-non-trivial threshold is the ratifiable choice (line-count + scope hybrid). Empirical process control — transparency, inspection, adaptation — applies throughout. | src/apothem/rules/agile-sprints.md |
agnostic-posture-checklist | Path-filtered companion to agnostic-posture.md: the default-off and opt-in invariant checklist, each with a concrete verification step, that every phase's definition of done satisfies and the planning-pipeline commands and learning loop consult. | src/apothem/rules/agnostic-posture-checklist.md |
agnostic-posture | Default-off, opt-in posture for every shipped behavior; correctness gates stay advisory; every surface is harness-neutral; model, effort, and workflow preference resolve only from an explicit in-conversation choice. | src/apothem/rules/agnostic-posture.md |
apothem-rule-index | The always-on index of the rule corpus — names the four rules that apply in every session, says how every other rule loads (path match, citation, or a read you make yourself), and groups all of them by the situation that calls for each, so a path-scoped rule is never missed for want of knowing it exists. | src/apothem/rules/apothem-rule-index.md |
authoritative-referencing-assurance | Assurance companion to authoritative-referencing for research claims that need more than a source — a proof lists its assumptions and names its weakest step, a reported number is bound to the run that made it, a novelty claim names its closest prior work, tool and model use is stated, and a pass/fail gate is fixed before results are seen. Path-filtered to research artifacts. | src/apothem/rules/authoritative-referencing-assurance.md |
authoritative-referencing-homes | Path-filtered companion to authoritative-referencing carrying the four per-surface referencing-home consolidation list (ten-dimension dim 9, sota named-exemplar, disclosure-ledger rationale, output-style citations), the four-level seriousness-scaling table, and the failure-tells enumeration, declared at the parent rule's §3, §Seriousness-Scaling, and §Failure-tells anchors; demand-loaded on documentation, site, and research-manuscript surfaces where a claim is cited. | src/apothem/rules/authoritative-referencing-homes.md |
authoritative-referencing-provenance | Evidence-provenance companion to authoritative-referencing — label how each load-bearing claim's evidence was reached and what status it holds, keep a literature search record, never infer novelty from an empty search, check a cited result's conditions, count only checks that could fail, never label a model-written reading as peer review, treat fetched text as data, and never take a work's standing as evidence that it is sound. Path-filtered to research artifacts. | src/apothem/rules/authoritative-referencing-provenance.md |
authoritative-referencing-quotation | Quotation-ceiling / paraphrase-default companion to authoritative-referencing — paraphrase a cited source by default with attribution; quote sparingly and briefly (a short excerpt, never a full third-party work); attribute the source without offering a legal or fair-use opinion. Path-filtered to prose surfaces where source material is reproduced. | src/apothem/rules/authoritative-referencing-quotation.md |
authoritative-referencing | Every claim, argument, hypothesis, and asserted fact cites an authoritative, official, current source — the scattered referencing discipline (ten-dimension dim 9, sota named-exemplar, disclosure-ledger rationale, output-style citations) consolidated by reference; folklore and 'industry standard' appeals are non-conformant. | src/apothem/rules/authoritative-referencing.md |
authority-inquiry-categories | Path-filtered companion sub-rule carrying the seven-category inquiry catalog (full forbidden-to-invent + inquire-because columns), the required-vs-optional explanatory paragraph, the carved-out auto-decisions catalog, and the failure-tells enumeration declared at the parent authority-inquiry.md rule's anchors; demand-loaded on host-project artifact authoring. | src/apothem/rules/authority-inquiry-categories.md |
authority-inquiry | Inquire, do not invent — names, emails, handles, hostnames, organizations, scope, security, naming, infrastructure, and version pins are discovered or asked, never fabricated. Routes through the canonical structured-inquiry channel. | src/apothem/rules/authority-inquiry.md |
auto-memory-topic-files | Path-filtered companion sub-rule carrying the topic-file management discipline, ecosystem-specific memory hygiene, session-end evaluation procedure, decision tree, and anti-patterns declared at the parent auto-memory.md rule's anchors; demand-loaded when the assistant edits any memory artifact. | src/apothem/rules/auto-memory-topic-files.md |
auto-memory | Auto memory lifecycle augmenting the harness built-in — classifies memory against PROGRESS.md / PLAN-NOTES.md / skills (stable fact/convention/preference/insight → memory; plan/task-specific → PROGRESS/PLAN-NOTES; reusable technique with detection signal → skill), bans session-specific context / incomplete info / CLAUDE.md duplicates / plan-specific decisions / ephemeral facts, and anchors topic-file hygiene to the path-filtered companion. | src/apothem/rules/auto-memory.md |
bidirectional-binding | Every substantive structural element carries reciprocal bindings to its peers in canonical five-direction notation — Drives → / Driven by ← / Satisfies → / Established by ↑ / Cross-bound with ↔. Every binding declared in one direction has a reciprocal back-pointer at the other end; half-edges are structural failures. A Bidirectional Binding Matrix appendix summarizes the graph where structure permits. | src/apothem/rules/bidirectional-binding.md |
canonical-layout-reporting-tiers | Path-filtered companion rule carrying the operational depth of rules/canonical-layout.md — the two-tier reporting body (per-sub-phase + phase rollup), the numeric-prefix discipline, the orphan-output recovery surface, the reciprocal producer / consumer cross-reference body, the surface-imposition table, and the M11 ↔ M12 correspondence table; demand-loaded on plan-suite / phase / migration touches. | src/apothem/rules/canonical-layout-reporting-tiers.md |
canonical-layout | Non-trivial multi-step work emits two-tier reporting (per-sub-phase report + phase-level rollup that aggregates rather than concatenates) and lays generated outputs at the host's discovered canonical layout with reciprocal producer / consumer cross-references and provenance. Orphan outputs (generated artifacts with no consumer / no index entry / no producer attribution) are structural failures. | src/apothem/rules/canonical-layout.md |
clean-architecture-layers | Clean architecture layer discipline — Domain / Application / Infrastructure / Presentation separation under the inward-only dependency rule (Domain depends on nothing; outer layers implement inner-layer interfaces injected at the composition root; no inner layer imports a concrete outer class). Mandatory at 3+ modules or SHARED+; relaxed for throwaway scripts and trivial scopes. Path-filtered to src / lib / app / packages source trees. | src/apothem/rules/clean-architecture-layers.md |
clean-room-generation-protocols | Path-filtered companion rule carrying the full Writing Protocol (§2), Re-Writing Protocol (§3), Code Generation (§4), Prose and Documentation (§5), Plan and Artifact Generation (§6), Decision Tree, and Anti-Patterns specifications anchored at the parent rules/clean-room-generation.md rule's pointer sections; demand-loaded when the assistant edits a governed-core authoring surface (CLAUDE.md, AGENTS.md, and rule, command, skill, agent, hook, and output-style Markdown files). | src/apothem/rules/clean-room-generation-protocols.md |
clean-room-generation | Clean-room generation methodology for every output — code, prose, plans, artifacts: each is a fresh derivation from an understood specification (independently re-derivable from its inputs), never a memorized template nor a cosmetic edit of existing content. Routes CM-4 search outcomes into the Writing vs. Re-Writing protocols and gates every re-write on quality elevation against a named deficiency. Implements CM-5 / CM-7 / CM-21. | src/apothem/rules/clean-room-generation.md |
code-craft-conventions | Universal code-craft delegation stub for languages without a dedicated per-language rule. Discovers the host's ratified formatter / linter / type-checker / test framework / per-language idioms and honors them per M1; surfaces silence as inquiry per M5; defers to a per-language sibling rule (code-craft-python, code-craft-shell, code-craft-markdown) when one matches the artifact's path. | src/apothem/rules/code-craft-conventions.md |
code-craft-markdown | Per-language code-craft for Markdown / prose artifacts — purpose-driven structure, sentence-level justification, precision over politeness, active-voice construction, hedge-elimination per the clean-room generation projection. Honors host's ratified Markdown linter (markdownlint, vale, prose-linter) and frontmatter conventions; passes the host's lint clean. | src/apothem/rules/code-craft-markdown.md |
code-craft-python | Per-language code-craft for Python — strict SOLID compliance, modern type hinting (3.10+ syntax, no Any), Google-style docstrings, specific exception handling, pytest discipline, security guardrails (no hardcoded secrets, no shell injection, no unsafe deserialization), and architectural anti-pattern interception. Path-filtered to Python source files and configuration manifests. | src/apothem/rules/code-craft-python.md |
code-craft-shell | Per-language code-craft for shell artifacts — POSIX bash idioms (set -euo pipefail; quoted variables; no eval on untrusted input; trap-based cleanup) and PowerShell idioms (Set-StrictMode -Version Latest; verb-noun cmdlet naming; no Invoke-Expression on untrusted input; structured error records). Honor shellcheck for bash and Invoke-ScriptAnalyzer for PowerShell; pass the host's ratified shell linter clean. | src/apothem/rules/code-craft-shell.md |
code-craft-ui | Path-filtered companion sub-rule to code-craft-conventions.md: the user-interface craft floor for component, markup and style files. Fixes the category priority order, a verify list and a refuse list for any UI change, and requires stated defaults in place of invented sourced values. | src/apothem/rules/code-craft-ui.md |
cognitive-identity-techniques | Path-filtered companion to rules/cognitive-identity.md carrying the detailed bodies of the five cognitive filters, the six ideation techniques (with detection signals and decision tree), the language-standards forbidden-phrase / required-quality / structural-format catalogs, and the five philosophical principles. Demand-loaded on substantive-output authoring surfaces. | src/apothem/rules/cognitive-identity-techniques.md |
cognitive-identity | The cognitive identity and creative architecture governing every substantive output — the cognitive-insurgent posture and Senior Software Architect role, the five sequential cognitive filters (obvious-purge through aesthetic-demand), the six on-detection ideation techniques, the forbidden / required language standards, and the seven-axes-of-breadth expertise taxonomy. Implements CM-21. | src/apothem/rules/cognitive-identity.md |
context-management-budget | Path-filtered companion rule carrying the §7 Context Budget Discipline (budget awareness, demand loading, pressure signals) and §7.4 Per-Task Effort Calibration (CM-12d) declared at the parent context-management.md rule's §7 anchor; demand-loaded on commands / PROGRESS / plan-suite artifact touches. | src/apothem/rules/context-management-budget.md |
context-management-protocol | Path-filtered companion rule carrying the detailed externalization sub-clauses, compaction-trigger catalog, long-conversation resilience procedures, graceful-degradation priorities, blind-execution protocol full body, and error-classification table declared at the parent context-management.md rule's §2/§3/§4/§5/§6/§8 anchors; demand-loaded on plan-workflow entry. | src/apothem/rules/context-management-protocol.md |
context-management-scratch | Path-filtered companion rule carrying the plan-workflow directory convention (_inputs/ working state, _spec/ authored specifications, and _outputs/ durable generated emissions) declared at the parent context-management.md rule's §2.6 anchor; demand-loaded on plan-workflow entry. | src/apothem/rules/context-management-scratch.md |
context-management | Systematic context management under the blind-execution invariant — every turn must be executable by a fresh session with zero prior history, so all state lives in durable files and active context is acceleration, not storage. Covers context-rot monitoring, proactive externalization, compaction discipline, opt-in continuous single-session execution, and context-budget calibration. Implements CM-12 / CM-14 / CM-18 / CM-19 / CM-24. | src/apothem/rules/context-management.md |
definitiveness-virtues | Path-filtered companion to rules/definitiveness.md carrying the operational depth of M8 — Definitiveness, Airtightness, and the Family of Rigorous-Systems Virtues. Demand-loaded when the assistant edits authoring surfaces (CLAUDE.md, AGENTS.md, rules, skills, agents, commands, hooks, output-styles) where prescriptive prose lands: the three hedge-resolution paths with worked examples, the seven-virtue floor, the definitive-form Right/Wrong examples, the airtightness checks, and the failure tells. | src/apothem/rules/definitiveness-virtues.md |
definitiveness | Every emitted statement is definitive and airtight — hedging vocabulary is eliminated where binding prescription is possible; every contract carries pre / post / failure conditions; every TBD / TODO / FIXME is closed in place or surfaced as an inquiry; the family of rigorous-systems virtues governs every artifact. | src/apothem/rules/definitiveness.md |
determinism | Every rendered option set carries the (Recommended) marker in its answer header, every terminal surface closes with a named next step, and every command, skill, output-style, and statusline surface holds a deterministic output shape — identical inputs produce identically-shaped output. The determinism harness proves the contract mechanically; the marker and next-step semantics are owned by their dedicated rules and this rule consolidates them under the determinism contract. | src/apothem/rules/determinism.md |
disclosure-ledger-markers | Path-filtered companion rule carrying the full marker-class enumeration, ledger-completeness detail, rationale-specificity detail, and failure-tells body declared at the parent disclosure-ledger.md rule's anchor; demand-loaded on disclosure-bearing artifact touches. | src/apothem/rules/disclosure-ledger-markers.md |
disclosure-ledger | Disclosed amendments, never silent — every change of meaningful scope carries an explicit ledger of what was asked, what was amended, what was extended, and what was deferred, with cited rationale. | src/apothem/rules/disclosure-ledger.md |
dynamism | No static substitutions for dynamic-source-of-truth values — every version, badge, release-reference, and docs-version surface renders dynamically from a single source of truth; static-string embeds across the 7 closed-set surfaces are structural failures. | src/apothem/rules/dynamism.md |
etc-extension | Enumerations are seeds, not ceilings — every 'etc.', 'e.g.', 'such as', 'like', 'including', and '…' is a directive to extend the set comprehensively from intent; only an enumeration explicitly marked closed is exhaustive. | src/apothem/rules/etc-extension.md |
expertise-posture-elements | Path-filtered companion rule carrying the seven sub-elements of expertise (full prose bodies), the four-rung calibration ladder, the disclosure marker enumeration, and the failure tells declared at the parent expertise-posture.md rule's anchors; demand-loaded on artifact-authoring touches. | src/apothem/rules/expertise-posture-elements.md |
expertise-posture | Read intent before literal text — amend proactively when literal request leaves a known defect; extend on adjacent gaps; refine with cited rationale; anticipate second-order consequences; calibrate depth to task; import lessons from adjacent domains. Never silent override of authority, agnosticism, or arrogance. | src/apothem/rules/expertise-posture.md |
freshness-facade | Shipped public surfaces carry no AI-disclosure, backward/legacy/obsolete/retired/deprecated, placeholder/TBD/coming-soon, or fix/refinement narrative — a current-version-only facade; the freshness-token-grep matcher gives the closed token-class list mechanical teeth on README plus site copy. | src/apothem/rules/freshness-facade.md |
harness-adapter-shape-schemas | Path-filtered companion to rules/harness-adapter-shape.md — carries the per-harness schema catalog, convergence-vs-divergence table, per-harness pre-emission gate adaptation, the 7-column adapter-test matrix template, the cross-harness redundancy hoist heuristic, and the per-harness STANDARD CONVENTION PIN templates for the 17-harness cohort. | src/apothem/rules/harness-adapter-shape-schemas.md |
harness-adapter-shape | Per-harness adapter discipline — discovery walk per harness, sibling-convention convergence, per-harness pre-emission gate adaptation, 7-column adapter-test matrix shape, cross-harness redundancy elimination, and the per-harness STANDARD CONVENTION PIN schema (vendor doc URL + commit-SHA + snapshot date + canonical filename + canonical schema). Demand-loaded on adapter-sub-package touches, co-triggering with its schemas companion. | src/apothem/rules/harness-adapter-shape.md |
host-discovery-manifests | Path-filtered companion rule carrying the per-language manifest catalog, the discovery-record provenance schema, the failure-tells enumeration, and the Derived Project Context Block definition declared at the parent host-discovery.md rule's §1, §4, Failure-tells, and §Derived Project Context Block anchors; demand-loaded on host-manifest or convention-document touches. | src/apothem/rules/host-discovery-manifests.md |
host-discovery | Discover host-project conventions before emitting any artifact — language, formatter, linter, layout, naming, idioms, sibling-file patterns. Honor discoveries; surface silence as an authoritative inquiry per the canonical channel. | src/apothem/rules/host-discovery.md |
i18n-discipline-locale-cohorts | Path-filtered companion sub-rule to i18n-discipline.md — carries the Modern Dev Cohort selection rationale + per-locale reach data, the RTL CSS-token catalog, and the hreflang link-tag shape + canonical-URL discipline. Demand-loaded on site / docs / locale-glossary / i18n-config touches. | src/apothem/rules/i18n-discipline-locale-cohorts.md |
i18n-discipline | Locale-cohort selection, framework i18n integration, machine-seed translation with optional human review, per-locale glossary, RTL discipline, and hreflang / lang attribute discipline for every host-project translated surface. | src/apothem/rules/i18n-discipline.md |
interactive-questions-canonical-shapes | Path-filtered companion rule carrying the canonical invocation shapes, worked examples, recommendation-taxonomy details, harness-fallback shape, destructive-op canonical option-set templates, and default-pointer worked examples for the parent interactive-questions.md rule; demand-loaded on path match. | src/apothem/rules/interactive-questions-canonical-shapes.md |
interactive-questions-detail | Path-filtered companion to interactive-questions.md carrying the authoring-discipline procedure and the anti-pattern catalog; demand-loaded on path match when a surface authors structured-inquiry invocations. | src/apothem/rules/interactive-questions-detail.md |
interactive-questions-sweep-matchers | Path-filtered companion rule carrying the canonical-channel sweep matcher specifications (H1–H7); demand-loaded when the parent interactive-questions.md rule's §8 anchor surfaces. | src/apothem/rules/interactive-questions-sweep-matchers.md |
interactive-questions | Canonical discipline for soliciting user input — the harness-neutral structured-inquiry channel is the sanctioned abstraction; every invocation carries a structured option annotation; destructive operations and ambiguity resolution route through this rule; a prose fallback shape covers harnesses that lack a native tool. | src/apothem/rules/interactive-questions.md |
large-file-generation | Large file generation via incremental appends — assess size (Small <200 / Medium 200–500 / Large 500–1500 / Massive >1500 lines), then for Large+ plan sections first and append via Write-then-Edit so documents are never silently truncated or abandoned. | src/apothem/rules/large-file-generation.md |
large-file-reading | Large file reading via targeted segments — Grep to locate, Glob for traversal, Read with offset/limit; never load a full file when a segment suffices. Read-side analogue of large-file-generation. Demand-loaded on file-consult touches; the always-on read-side budget invariant lives in context-management.md §7. | src/apothem/rules/large-file-reading.md |
living-docs | Every change to a documented public surface — a CLI command or flag, a harness adapter, a profile field, an installer flag or environment variable, a configuration key, or any surface with a page under site/content/docs/ — MUST update its documentation page in the same change-set; source-generated reference pages are kept current by re-running the reference-inventory generator and committing the regenerated pages. | src/apothem/rules/living-docs.md |
multi-agent-workflow-shape | Path-filtered companion sub-rule to multi-agent-workflow.md — carries the four-property capability-shape enumeration (independent agents, adversarial critique, open-loop dynamics, dynamic pacing), the model- and effort-agnostic detail, the synthesis-quality-and-lean-main-thread form, and the seriousness-scaling recommendation table declared at the parent rule's §1, §3, and §Seriousness-Scaling anchors; demand-loaded on agent / workflow-command / workflow-skill and multi-agent authoring surfaces. | src/apothem/rules/multi-agent-workflow-shape.md |
multi-agent-workflow | Independent-critique, open-loop, dynamic multi-agent execution is an available, specified capability — opt-in and default-off under the agnostic posture, never imposed on a clean install; the orchestration mechanics are owned by agent-orchestration. | src/apothem/rules/multi-agent-workflow.md |
operational-mandates-expanded | Path-filtered companion to operational-mandates — carries the per-mandate Violation indicators and Recovery sub-blocks for CM-1 through CM-10; demand-loaded on Markdown / governed-core surface touches. | src/apothem/rules/operational-mandates-expanded.md |
operational-mandates | The ten behavioral mandates CM-1–CM-10 as canonical one-line directives — critical evaluation, zero assumptions, configuration-driven, search-before-implement, best solution, self-improvement, coherent product, bottleneck-first focus, decision velocity, and brutal honesty. The behavioral floor every other rule reads through; per-mandate violation indicators and recovery actions live at the path-filtered companion. | src/apothem/rules/operational-mandates.md |
option-annotation-form | Path-filtered companion rule carrying the prose-and-document annotation form, rationale-specificity citations, cross-channel consistency invariant, cardinality bounds, and failure-tells enumeration declared at the parent option-annotation.md rule's anchor lines; demand-loaded on prose-and-document artifact touches. | src/apothem/rules/option-annotation-form.md |
option-annotation | Every multi-option choice carries a Recommended marker plus principle-linked rationale — silent picks and un-annotated option lists are forbidden in authoritative territory. The structured-inquiry subset delegates to the canonical structured inquiry schema; this rule extends the discipline to prose-and-document option sets. | src/apothem/rules/option-annotation.md |
own-voice-reimplementation | Every reference-derived feature is reauthored in apothem's own voice with zero verbatim code/text/branding, is demonstrably more elegant and effective than the reference, and cites the relevant harness's own latest official documentation (a convention-pin snapshot dated within the prior 90 days). | src/apothem/rules/own-voice-reimplementation.md |
performance-discipline | Per-class runtime performance budgets with quantitative pass/fail gates — each artifact class (hook handlers, the conformity sweep, the test suite, agent spawn, install resolution) carries a runtime ceiling, a measurement boundary, and a benchmark verifier that exits zero on compliance; budget exceedances surface as findings routed to the expertise-gap log. Closes the Performance axis of the seven-axes-of-breadth taxonomy. | src/apothem/rules/performance-discipline.md |
persistent-conventions-vigilance-checklist | Path-filtered companion sub-rule carrying the Convention Awareness checklist (§1) and the Ecosystem Gap Detection artifact-class triggers (§4) for the parent persistent-conventions-vigilance.md rule; demand-loaded on touches of governed-core surfaces. | src/apothem/rules/persistent-conventions-vigilance-checklist.md |
persistent-conventions-vigilance | Adhere to the harness ecosystem's conventions and proactively evolve its artifacts — per-artifact-class convention awareness (naming, layout, frontmatter shape), the five-step proactive-evolution cycle (detect → search → act → validate → retire), bidirectional cross-reference coherence, and ecosystem gap detection that surfaces a missing rule / skill / command / hook before it is needed. Implements CM-22. | src/apothem/rules/persistent-conventions-vigilance.md |
plain-language | User-facing apothem narrative reads as human-authored with zero process-tooling leak — a closed-set forbidden vocabulary (AI / agent / LLM / attestation / ratified / cutover-rehearsal / harness brand identifiers / plan-stage tokens) is eliminated from in-scope codebase-artifact surfaces; the product noun harness is a domain carve-out; dev-facing tech-spec, rule-tier, and per-harness materializer modules are out-of-scope process artifacts that retain their native vocabulary. | src/apothem/rules/plain-language.md |
planning-techniques | Nine planning-review techniques — each pairing a Detect signal, a verification Procedure, and the Anti-pattern it guards against: iteration-loop safety, campaign dependency ordering, clean-slate artifact conflict, identity-claim verification, asymmetric-fix propagation, blind-review value, added-task completeness, severity-count verification, and strict-gate remediation. Applied on detection across plan generation, review, audit, and execution. Implements CM-20 / CM-21. | src/apothem/rules/planning-techniques.md |
pre-emission-gate-bars | Path-filtered companion rule carrying the full Fifteen-Bars table (M1–M15 detailed Check + Failure→action columns), the Attestation Schema YAML block, and the iteration-on-failure protocol declared at the parent pre-emission-gate.md rule's anchor; demand-loaded when the assistant edits any artifact whose emission triggers the fifteen-bar gate. | src/apothem/rules/pre-emission-gate-bars.md |
pre-emission-gate | Fifteen-bar pre-emission gate — every host-project artifact passes M1 through M15 before emission, with a recorded YAML attestation and explicit n/a reasoning where a bar does not apply. | src/apothem/rules/pre-emission-gate.md |
production-ready-prs-surfaces | Path-filtered companion to rules/production-ready-prs.md carrying the operational depth of M15: the seven visibility surfaces, the supply-chain posture catalog, the release-engineering invariants, the commit-message convention (with the human-only authorship clause), the CI-green discipline, and the modern-project-surface specification. Demand-loaded when the assistant edits any visibility-surface artifact (README / CHANGELOG / CONTRIBUTING / LICENSE / SECURITY / .github/** / install / update / uninstall scripts). | src/apothem/rules/production-ready-prs-surfaces.md |
production-ready-prs | Every change the agent makes in a host project is delivered in production-ready form — tests + docs + CHANGELOG entry + conformant commit message + CI green in the same change-set. The seven visibility surfaces (what-is-this / how-to-install / is-it-alive / is-it-safe / how-to-contribute / can-I-trust / what-changed) are honored; gaps surface as findings rather than silently entrenched. | src/apothem/rules/production-ready-prs.md |
propagation | Every mutation of any artifact propagates, in the SAME change-set, to every dependent reference across the whole repository's reference graph — code, tests, docs, root files, .github, CHANGELOG, harness-rendered templates, the harness registry, plugin manifests, and cross-rule bindings. A mutation (add / edit / remove / rename / move / split / merge / deprecate) that lands without its reference updates is an orphan / half-edge finding. Unifies the docs-tier (living-docs) and component-tier (systemic-participation) propagation disciplines to the full reference graph; the existing drift gates are its mechanical arm. | src/apothem/rules/propagation.md |
recommend-next-step | Every command, skill, and phase artifact closes with a Recommended Next Step block — canonical heading, named action, optional rationale — so terminal output never leaves the operator without a definitive forward move. Hedged or absent next-step blocks are structural failures; the mechanical matcher operationalises the check at the pre-emission gate. | src/apothem/rules/recommend-next-step.md |
refactoring-discipline | Agent-driven refactoring is test-gated, one-at-a-time, plan-first, and continuous: a behavior-preserving refactor holds a GREEN test baseline before the first edit and after (the before-and-after contract); addresses exactly one concern in an isolated workspace; proceeds only from a reviewed plan after maximal context-gathering; and runs continuously as the codebase drifts rather than deferred to a crisis. Demand-loaded on refactor-class source edits; the detection signals and full procedure live in the body. | src/apothem/rules/refactoring-discipline.md |
session-closure-scaling | Path-filtered companion sub-rule to session-closure.md — carries the universality-and-scaling clause (the close binds every session; depth scales to session weight), the single-canonical-close no-repetition discipline (emit once, report only the delta on re-engagement), and the plan-touching residual detail (plan open items zeroed or explained-or-waived on every plan-touching close). Demand-loaded on plan-suite / context-management / recommend-next-step touches. | src/apothem/rules/session-closure-scaling.md |
session-closure | Every session — ad-hoc conversational sessions included, not only plan phases — ends with a formal, verifiable close: a terminal Recommended Next Step, a done/deferred ledger, and a verification attestation. A session that trails off mid-thread, claims done without a checked outcome, or buries deferred work silently is a structural failure. Harness-agnostic; the close is rules text every harness honors. | src/apothem/rules/session-closure.md |
sota-elevation-exemplars | Path-filtered companion to sota-elevation.md carrying the eight-surface SOTA evaluation lens, the named-exemplar catalogue (cpython / shadcn-ui / Bun / Fumadocs / vite / vitest), per-surface exemplar walks, the upper-bound calibration binding, the Filter-5 aesthetic-demand binding, the gap-surfacing protocol, the disclosure markers, and the failure tells; demand-loaded on README / docs / .github/ / site/ touches. | src/apothem/rules/sota-elevation-exemplars.md |
sota-elevation | SOTA elevation as default posture for OSS distribution projects — every user-facing surface targets the upper bound of contemporary best-practice, calibrated against named exemplar projects (cpython, shadcn-ui, Bun, Fumadocs, vite, vitest) and audited across the eight SOTA evaluation surfaces. | src/apothem/rules/sota-elevation.md |
source-accessibility-scaling-tells | Path-filtered companion to source-accessibility carrying the four-level seriousness-scaling table and the failure-tells enumeration for source-trust-outranks-reachability selection, declared at the parent rule's §Seriousness-Scaling and §Failure-tells anchors; demand-loaded on documentation, site, and research-manuscript surfaces where a claim's source is selected and cited. | src/apothem/rules/source-accessibility-scaling-tells.md |
source-accessibility | Source trust outranks source accessibility — reach a trusted but inaccessible source via browser then operator interview; never prefer an untrusted-but-free source over a trusted-but-inaccessible one; record the source-trust decision in the ledger. | src/apothem/rules/source-accessibility.md |
surgical-manipulation | Every mutation of an existing artifact is surgical — a precise, minimal, anchor-bounded or managed-block edit that touches only the span the change requires, never a blunt whole-file overwrite where a scoped edit suffices, and never a strip of a surrounding banner or unrelated content. The mutation produces a minimal diff attested against the host's green conformity baseline (the golden-corpus invariant). The surgical-edit + reactive-guard mechanism is the surgical-guard skill. | src/apothem/rules/surgical-manipulation.md |
systemic-participation-relations | Path-filtered companion rule carrying the operational detail of M14 systemic participation — the four systemic relations table, sibling-convention convergence detail, index/registry update table, the three orphan classes, the three silo classes, retirement discipline, and the cross-component reference graph invariant. Demand-loaded when the parent systemic-participation.md rule's anchors surface on host-project component touches. | src/apothem/rules/systemic-participation-relations.md |
systemic-participation | Every artifact apothem introduces into a host project joins the host as a systemic participant — declares its upstream / downstream / peers / enforcers; honors sibling conventions; lands at the host's canonical layout; updates the host's index registries in the same change. Silos (self-contained artifacts diverging from sibling conventions) and orphans (artifacts with no consumer / no index entry / no producer attribution) are structural failures, not aesthetic preferences. | src/apothem/rules/systemic-participation.md |
ten-dimension-check-dimensions | Path-filtered companion sub-rule carrying the verbatim per-dimension bodies (rigor / coherence / configurability / readability / orphanism / structurality / architecture / naming / scholarly referencing / examples-tests-docs) and per-dimension failure tells declared at the parent ten-dimension-check.md rule's anchor; demand-loaded on artifact-emission surfaces. | src/apothem/rules/ten-dimension-check-dimensions.md |
ten-dimension-check | Ten quality dimensions every host-project artifact passes before emission — rigor, coherence, configurability, readability, orphanism, structurality, architecture, naming, scholarly referencing, examples / tests / docs. | src/apothem/rules/ten-dimension-check.md |
token-budget-discipline | Caps every always-on rule body at 500 substantive tokens and the whole always-on set at a byte ceiling (four core rules plus the rule index); over-budget bodies decompose into path-filtered companion sub-rules per the established demand-load pattern; mechanical matcher and PreToolUse hook enforce the budget at write time. | src/apothem/rules/token-budget-discipline.md |
token-efficiency-rewrite-protocol | The full token-efficiency rewrite protocol the always-on parent rule references: the L1 prose-scaffolding elimination classes (filler / throat-clearing / restatement / hedge-padding / ceremonial-scaffolding / connective-bloat), the per-class compression heuristics, the extract-discard-re-derive-verify procedure, and worked before/after examples. Demand-loads when the assistant rewrites an authoring surface for token efficiency. | src/apothem/rules/token-efficiency-rewrite-protocol.md |
token-efficiency-rewrite | Rewrite protocol for token-efficient content — re-derives prose to a minimal surface while preserving L2 substantive semantics and L3 structural / load-bearing anchors; pairs with token-budget-discipline (sizing) and clean-room-generation §3 (re-writing protocol). | src/apothem/rules/token-efficiency-rewrite.md |
tool-use-discipline-failure-tells | Path-filtered companion to tool-use-discipline carrying the failure-tells enumeration for the three tool-loop disciplines — sequential-where-parallel, act-without-observe, no-verifiable-exit, fixed-count-stop, and false-sequencing — declared at the parent rule's §Failure-tells anchor; demand-loaded on rule, command, skill, agent, hook, and output-style authoring surfaces where tool-loop cadence is exercised. | src/apothem/rules/tool-use-discipline-failure-tells.md |
tool-use-discipline | Ordinary tool use runs as a disciplined loop: independent tool calls go in one turn, never sequentially; the agent works an observe → decide → act cadence; the loop iterates until a verifiable exit condition is met, never a fixed count. Generalizes the agent-tier single-message parallel-launch discipline down to every tool call. Harness-agnostic; advisory under the agnostic posture. | src/apothem/rules/tool-use-discipline.md |
visual-leverage | Structural subject matter — architecture, control flow, data flow, dependency graph, state machine, sequence, decision tree, hierarchy, precedence stack, lifecycle, permission matrix — carries a current-reality diagram alongside its prose, with provenance and a verification date. Mermaid is the recommended default for Markdown-centric ecosystems; the host's existing notation is honored per M1 host-discovery. | src/apothem/rules/visual-leverage.md |