Agent Governance Framework
This document is the canonical governance policy for agent-facing instructions in this repository. It remains adocs-guide/policies/ page, not a docs-guide/frameworks/ page, because canonical agent-file locations, overlap boundaries, and retirement targets are enforced governance decisions rather than optional improvement guidance.
The approved governance model is:
- one cross-agent baseline
- one native adapter per supported agent family
- one shared repo knowledge set
- no policy drift between overlapping instruction files
Approved Canonical Layout
Approved target locations:
Current implementation state:
Governance Taxonomy
Must-have canonical rules
These rules must live in the canonical cross-agent baseline and must not drift by agent:- destructive git safety
- hook and validation policy
- local preview port policy (
3000forbidden for agent-run Mintlify/preview sessions) - root allowlist policy
- source-of-truth order
v1freeze / legacy content boundary- branch and lock governance
- path-migration discipline
Recommended shared rules
These rules should be shared across agents unless a native adapter requires different syntax:- response and review hygiene
- component and style conventions
- route-placement heuristics
- change-propagation expectations
- generated-file handling
- contributor workflow norms
Native adapter rules
Native adapter files should contain only:- tool-specific file format or metadata requirements
- minimal bootstrap instructions for that tool
- a pointer back to the canonical repo baseline
Required Repo Knowledge for Every Agent
Every supported agent must be grounded in the same repo knowledge set:Current Active Rule Inventory and Overlap Map
Retired Legacy Surfaces
These files were intentionally retired during the approved governance rollout and should not be recreated as active governance sources:Overlap clusters
Resolved in Phase 9:.github/AGENTS.mdnow carries disambiguating HTML comment; no longer a duplicate of rootAGENTS.mdASSISTANT.mdis a redirect stub; canonical content at.mintlify/Assistant.md.github/augment-instructions.mdis archived; replaced by.augment/rules/
docs-guide/contributing/agent-instructions.mdxstill contains contributor-facing parts of the policy set; reduce to a link to AGENTS.md in a future passai-tools/agent-packs/*are generated from separate sources and must remain outputs, not canonical rule inputs
Approved Migration Map
The current governance contract maps legacy surfaces to the following destinations:Prior Reports Incorporated
This policy incorporates the following prior reports and decisions:Prior report: Root entry re-review (2026-03-16)
Relevant preserved findings:
- not everything currently discussed in
.allowlistis required at repo root - nested agent paths such as
.github/copilot-instructions.mdare not valid.allowlistentries because the root gate is root-only - the current repo still uses
ASSISTANT.md, and legacy assistant surfaces were identified as move or retire candidates
Prior report: Agent root file decision set (2026-03-16)
Approved decision set carried forward into this policy:
- Codex should use
AGENTS.mdas the cross-repo baseline;.github/AGENTS.mdadds Codex-specific task-isolation rules - GitHub Copilot should use
.github/copilot-instructions.md - Claude Code should use
.claude/CLAUDE.md - Cursor should use
.cursor/rules/*.mdc - Windsurf should use
.windsurf/rules/*.md - Augment Code should use
.augment/rules/*.md - Mintlify chat widget reads
.mintlify/Assistant.md— this is not an AI coding assistant adapter
Report Sources
Repo evidence
.github/AGENTS.md.github/copilot-instructions.md.github/augment-instructions.mdASSISTANT.mddocs-guide/contributing/agent-instructions.mdxai-tools/agent-packs/README.mdai-tools/agent-packs/claude/CLAUDE.mdai-tools/agent-packs/cursor/rules.mdai-tools/agent-packs/windsurf/rules.md.codex/README.md.codex/task-contract.yamloperations/scripts/validators/governance/check-agent-docs-freshness.jsdocs-guide/tooling/ai-tools.mdxdocs-guide/policies/root-allowlist-governance.mdx