aidd-policy

Scope

• This skill is the single source of truth for reusable policy guidance.
• Stage/runtime ownership stays in feature-dev-aidd:aidd-core and stage skills.
• Preload matrix v2: this skill is required for every subagent role.

Output contract (required)

• Status: ...
• Work item key: ...
• Artifacts updated: ...
• Tests: ...
• Blockers/Handoff: ...
• Next actions: ...
• AIDD:READ_LOG: ...
• AIDD:ACTIONS_LOG: ...

Include Checkbox updated: ... when the command/stage contract expects it.

Question format

Use this exact format when asking the user:

Question N (Blocker|Clarification): ...
Why: ...
Options: A) ... B) ...
Default: ...

Read policy (pack-first)

• Read packs/excerpts before full documents.
• Prefer rlm_slice.py and section slices for focused evidence.
• Full-file reads are allowed only when slices are insufficient.
• Keep AIDD:READ_LOG compact and point to aidd/reports/**.

Loop safety

• Loop stages run through canonical stage-chain (preflight -> run -> postflight -> stage_result).
• Manual stage-result writes remain forbidden operator paths.
• Missing preflight/readmap/writemap/actions/log artifacts is a contract violation.

Runtime path safety

• Never guess runtime filenames/paths.
• Execute only runtime entrypoints explicitly listed in the current stage skill Command contracts.
• Use only canonical active-state runtime commands from the current stage contract; do not use deprecated aliases.
• For loop stages, internal preflight runtime remains a stage-chain orchestration detail, not a user-facing command contract.
• If command output contains can't open file .../skills/.../runtime/..., stop and report BLOCKED (runtime_path_missing_or_drift) with evidence path.

Retry safety

• Do not repeat the same failing shell command in a loop.
• For cwd/build-tool mismatch (./gradlew missing, or does not contain a Gradle build), allow at most one corrected attempt after explicit cwd resolution.
• If corrected attempt still fails, stop with blocker (tests_cwd_mismatch) and handoff instead of further retries.

Subagent guard

• Subagents must not edit aidd/docs/.active.json.

Command contracts

Policy output contract application

• When to run: before final response for every stage/subagent output.
• Inputs: current stage result plus artifacts/tests/blockers evidence.
• Outputs: deterministic response skeleton with required contract fields.
• Failure mode: missing required fields, ambiguous status, or inconsistent blocker/handoff reporting.
• Next action: normalize output structure and rerun final response composition.

Policy question protocol

• When to run: only when blocker/clarification is unavoidable after artifact-first checks.
• Inputs: verified blocker context, user-facing options, and default path.
• Outputs: one compact policy-compliant question using Question N/Why/Options/Default format.
• Failure mode: free-form or multi-topic questions that lack options/default.
• Next action: rewrite the question into deterministic policy format before asking.

Additional resources

• references/output-contract.md (when: response structure is unclear; why: copy exact contract fields and stage notes).
• references/read-policy.md (when: deciding between slice and full read; why: keep evidence gathering deterministic and minimal).
• references/question-format.md (when: blockers require user clarification; why: ask short, actionable questions with defaults).
• references/loop-safety.md (when: loop-stage safety flags or artifacts are in question; why: apply consistent blocked/warn policy).