Repository agent instructions

Default BMAD workflow

BMAD Method 6.11.0 is installed from the locked bmad-method package. Bundled core is present, and BMM is the only selected optional module. BMAD is the repository default for qualifying future extension planning and implementation. Before planning or implementation, read _bmad-output/project-context.md. Register durable work in _bmad-output/index.yaml and resolve human approval through _bmad-output/authorities.yaml. Run scripts/verify-bmad.sh before completion.

Tool entrypoint

For Claude Code, use the tracked .claude/skills/bmad-* launchers with equivalent guidance. Use bmad-help when routing is unclear.

Repository authority and workflow sources

Cargo.toml, ARCHITECTURE.md, SECURITY.md, studies/rust-agent-v1/README.md, studies/rust-agent-pilot/README.md, docs/book.toml, docs/src/SUMMARY.md, and .github/workflows/ci.yml are authoritative over generated BMAD guidance. Apply each within its documented scope. If authoritative sources appear to conflict, stop and request repository-maintainer review; generated guidance must not choose between them.

Use _bmad/_config/manifest.yaml as the installation manifest and _bmad/_config/skill-manifest.csv as the canonical source for skill and workflow identities and generation-provenance source paths. Paths in the skill and file manifests are not deployed _bmad runtime targets. Each tracked bmad-*/SKILL.md under both launcher roots is a runtime tool entrypoint. Only when an entrypoint explicitly renders or resolves a workflow, follow the single workflow path it returns.

Workflow tiers

Apply these tiers in order without reordering or skipping a higher tier:

1. Protected-study prohibition. It has no exemption.
2. Urgent security procedure. Create a minimal incident record approved by an eligible security-owner or Don from the authority registry. The retrospective deadline is the next five Monday-through-Friday calendar days in Europe/London, with no holiday exclusions. After validation against that rule, the recorded deadline is authoritative. The incident record is temporary routing; a retrospective full-BMM artifact is required by that recorded deadline.
3. Mandatory full BMM planning for features, behavioral changes, architectural changes, public APIs, dependency changes, security-sensitive changes, and cross-crate work.
4. Trivial exemption only for typo-only, formatting-only, or mechanically generated-only changes when no mandatory category applies. Record a rationale and obtain approval from an eligible repository-maintainer who is not the author, unless the registry's Don explicitly approves and evidences the independence exception.

Information classification is human-reviewed only. The allowed values are public and repository-visible. Repository-visible means visible to everyone with repository access and excluded from public documentation by default; it is not a confidentiality or access-control boundary.

Protected studies

BMAD must not mutate or use studies/rust-agent-v1 or studies/rust-agent-pilot as scratch space. Automatically apply the same protection to every future study whose protocol declares itself frozen, locked, or pre-registered. A change to any protected task, grader, prompt, reference patch, model, effort, image, budget, retry policy, endpoint, publication gate, or other frozen term requires a new study.

Data and publication boundaries

Do not place credentials, secrets, personal data, customer-confidential information, live exploit material, raw private evidence, or anything needing restricted storage in Git or BMAD output.

_bmad and _bmad-output are internal by default. Do not add them to public mdBook navigation or links, documentation copy rules, or rendered output unless publication is explicitly reviewed. docs/book.toml and docs/src/SUMMARY.md define the public mdBook boundary.

Upgrade policy

Never run the installer or regeneration in the working checkout. An exact stable patch update requires renewed provenance review, regeneration and comparison in a disposable clean clone, generated-only promotion, human review, and full verification. Any minor or major update requires a new approved plan. Never overwrite human-owned governance, project context, durable artifacts, or shared configuration.

Machine policy

The following marker-delimited JSON is the canonical cross-tool policy mirror. Keep the complete marker, JSON, and marker bytes identical in AGENTS.md and CLAUDE.md.

{ "artifact_governance": { "allowed_classifications": [ "public", "repository-visible" ], "authorities": "_bmad-output/authorities.yaml", "classification_decision": "human-reviewed-only", "index": "_bmad-output/index.yaml", "repository_visible_meaning": "visible-to-all-repository-readers-not-a-confidentiality-or-access-control-boundary" }, "authoritative_sources": { "conflict_action": "stop-and-request-repository-maintainer-review", "paths": [ "Cargo.toml", "ARCHITECTURE.md", "SECURITY.md", "studies/rust-agent-v1/README.md", "studies/rust-agent-pilot/README.md", "docs/book.toml", "docs/src/SUMMARY.md", ".github/workflows/ci.yml" ], "precedence": "repository-authorities-override-generated-bmad-guidance" }, "bmad": { "default_intent": "qualifying-future-extension-planning-and-implementation", "installation_manifest": "_bmad/_config/manifest.yaml", "installer": { "locked": true, "name": "bmad-method", "version": "6.11.0" }, "lock": "bmad.lock.json", "method": "BMAD Method", "module": { "id": "bmm", "name": "BMad Method", "only_selected_optional_module": true, "version": "6.11.0" }, "workflow_manifest": "_bmad/_config/skill-manifest.csv", "workflow_source": { "entrypoint_filename": "SKILL.md", "entrypoint_pattern": "bmad-*/SKILL.md", "entrypoint_role": "tool-entrypoint", "identity_and_path_source": "_bmad/_config/skill-manifest.csv", "manifest_path_semantics": "generation-provenance-not-deployed-runtime-targets", "resolution_rule": "when-entrypoint-explicitly-renders-or-resolves-a-workflow-follow-the-single-workflow-path-it-returns" } }, "forbidden_data": [ "credentials", "secrets", "personal-data", "customer-confidential-information", "live-exploit-material", "raw-private-evidence", "restricted-storage-material" ], "launchers": { "claude-code": { "prefix": "bmad-", "root": ".claude/skills" }, "codex": { "prefix": "bmad-", "root": ".agents/skills" } }, "policy_schema_version": 1, "project_context": "_bmad-output/project-context.md", "publication": { "forbidden_without_review": [ "navigation", "links", "copy-rules", "rendered-output" ], "internal_roots": [ "_bmad", "_bmad-output" ], "public_book_config": "docs/book.toml", "public_navigation": "docs/src/SUMMARY.md", "public_source_root": "docs/src", "review_required": true }, "studies": { "current_protected_roots": [ "studies/rust-agent-v1", "studies/rust-agent-pilot" ], "future_marker_keywords": [ "frozen", "locked", "pre-registered" ], "future_marker_rule": "protocol-declaration-automatically-protects-study", "mutation_allowed": false, "new_study_required_for_changes_to": [ "task", "grader", "prompt", "reference-patch", "model", "effort", "image", "budget", "retry-policy", "endpoint", "publication-gate", "other-frozen-term" ], "scratch_space_allowed": false }, "tiers": [ { "exemption_allowed": false, "id": "protected-study-prohibition", "order": 1, "rule": "protected-studies-cannot-be-mutated-or-used-as-scratch-space" }, { "approver_roles": [ "security-owner", "Don" ], "id": "urgent-security-procedure", "order": 2, "record": "minimal-incident-record", "retrospective_deadline": { "business_days": 5, "calculation": "next-five-Monday-through-Friday-calendar-days", "holiday_exclusions": false, "recorded_deadline_authoritative_after_validation": true, "timezone": "Europe/London" }, "retrospective_full_bmm_artifact_due_by_recorded_deadline": true, "temporary_routing": true }, { "categories": [ "features", "behavioral-changes", "architectural-changes", "public-apis", "dependency-changes", "security-sensitive-changes", "cross-crate-work" ], "id": "mandatory-full-bmm-planning", "order": 3, "planning": "full-BMM" }, { "approver_independent_of_author": true, "approver_role": "repository-maintainer", "don_exception": { "allowed": true, "evidence_required": true, "explicit_approval_required": true }, "eligible_changes": [ "typo-only", "formatting-only", "mechanically-generated-only" ], "id": "trivial-exemption", "mandatory_category_must_not_apply": true, "order": 4, "rationale_required": true } ], "upgrades": { "human_owned_paths_must_not_be_overwritten": [ "_bmad-output/authorities.yaml", "_bmad-output/implementation-artifacts", "_bmad-output/index.yaml", "_bmad-output/planning-artifacts", "_bmad-output/project-context.md", "_bmad/custom/config.toml" ], "major_update_requires_new_approved_plan": true, "minor_update_requires_new_approved_plan": true, "stable_patch_update_must_be_exact": true, "stable_patch_update_steps": [ "renewed-provenance-review", "disposable-clean-clone-regeneration-and-comparison", "generated-only-promotion", "human-review", "full-verification" ], "working_checkout_installer_or_regeneration_allowed": false }, "verifier": "scripts/verify-bmad.sh" }