pharaoh-audit-fanout

pharaoh-audit-fanout

When to use

Invoke at the start of a full project audit, CI gate, or pre-release review. Produces a deduplicated, terminology-consistent findings report covering 5 sub-areas in parallel.

Compositional design (not atomic by design)

This is an orchestrator skill. Atomicity criteria (a) does not apply — the skill is compositional by construction. Its 5 component skills each pass (a)-(e) individually.

Sub-areas

#	Atomic skill	Target issues
1	pharaoh-coverage-gap	orphans, unverified req, uncovered req-verification
2	pharaoh-lifecycle-check	invalid state transitions, stale reviews
3	pharaoh-standard-conformance	ISO 26262 §6 / ASPICE indicator gaps
4	pharaoh-review-completeness	missing reviewer / approved_by fields
5	pharaoh-process-audit	duplicate req, contradictory pair, missing FMEA, broken back-link

Shared memory protocol

All 5 subagents write findings to the project's .papyrus/ workspace using pharaoh-finding-record. The deterministic-ID scheme in pharaoh-finding-record provides natural dedup via Papyrus's FileLock-guarded ID-collision semantics.

Subagents SHOULD NOT retry on action: duplicate — that's the intended signal that another subagent already covered this issue.

Process

Step 1: Initialize workspace

Ensure <project_dir>/.papyrus/ exists and is writable. If not, initialize via papyrus --workspace <project_dir>/.papyrus init (zero-config, silos preset).

Step 2: Dispatch 5 subagents in parallel

In a production Pharaoh deployment, dispatch via Agent tool with subagent_type: general-purpose. In the Phase 4b harness, parallelism is achieved at the harness level via concurrent.futures.ThreadPoolExecutor over 5 separate claude -p invocations, each primed with one sub-area's task.

Each subagent receives:

• The project directory path
• Its assigned sub-area skill name (one of the 5)
• Explicit instruction to invoke pharaoh-finding-record for every finding

Step 3: Aggregate

Once all 5 subagents complete, read the Papyrus workspace via:

papyrus --workspace <project_dir>/.papyrus recall --tag category:* --format full

Emit as the final audit report (JSON list of findings, ordered by category).

Input / output

Input: {project_dir: path, scope?: "full"|"partial"}. Default scope is full (all 5 sub-areas). Partial scope limits to a subset of sub-areas (not used in Phase 4b testing).

Output: JSON list of aggregated findings, one object per (category, subject_id) pair:

[
  {"category": "orphan_arch", "subject_id": "arch__orphan_0", "finding_text": "...", "reporter_id": "coverage-gap"},
  ...
]

Failure modes

• .papyrus/ workspace uninitializable → fall back to sequential single-agent pharaoh-process-audit with stderr warning.
• Any subagent errors → other 4 continue; final report notes the missing sub-area.

• 1 subagent timeout → abort and surface partial Papyrus state.

Composition

Consumed by CI gates, on-demand user audits, and release governance workflows. Output is directly consumable by pharaoh-audit-report (future skill) or printed as-is.