Skip to content

Bound predict-audit ownership-walk + rule-sweep harnesses; consolidate unverified findings#1088

Merged
0xaslan merged 5 commits into
mainfrom
at/predict-audit-bound-siblings
Jun 25, 2026
Merged

Bound predict-audit ownership-walk + rule-sweep harnesses; consolidate unverified findings#1088
0xaslan merged 5 commits into
mainfrom
at/predict-audit-bound-siblings

Conversation

@0xaslan

@0xaslan 0xaslan commented Jun 25, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • Bound the ownership-walk and rule-sweep audit harnesses by construction: MAX_ROUNDS default 12→3 (ownership-walk) and 10→3 (rule-sweep), add verifyCap (60), and severity-gate verify (cleanup-tier findings reported raw, no subagent). Total agents ≤ maxRounds×units + verifyCap, so neither harness can run away to the agent cap.
  • Each sibling harness now emits an unverified array (Info/Low/cleanup findings triaged out of the verify panel).
  • consolidate.py now records those unverified findings — under their own key in findings.json (so track.py doesn't flood the tracker with low-priority noise), kept in the DROPPED 0 accounting (no silent drop), and surfaced in a dedicated Unverified report section.
  • Repo hygiene: untrack docs/superpowers/specs/2026-06-23-propbook-block-scholes-feed-split-design.md. It was committed long ago (219a9103) and stayed tracked, so the /docs/ and /docs/superpowers/ .gitignore rules never applied — git skips ignore rules for already-tracked files, which is why these specs kept reappearing in PRs. git rm --cached untracks it (local copy kept); the existing rule now keeps it (and future superpowers specs) out.

Key decisions

  • This is the sibling half of the predict-audit harness rework; the orchestrator + SKILL.md half already landed in main via the reference-tick admission PR. The orchestrator-only features (per-lane retirement, cross-model codex verify, depth tiers) are intentionally not ported to the siblings here — that's a separate follow-up.
  • The consolidate.py change re-syncs the consolidator with the already-merged orchestrator (which emits unverified), closing a split where the orchestrator landed without its matching consolidator handler.
  • Read-only on packages/* — no smart-contract changes; audit tooling only.

Test plan

  • python3 .claude/skills/predict-audit/evals/test_consolidate_track.py✅ ALL PASS (locks no-slip / id-dedup / tracker invariants; required after editing consolidate.py)
  • node --check (runtime-wrapped) parses ownership-walk.workflow.js + rule-sweep.workflow.js
  • git check-ignore -v confirms the untracked spec now matches /docs/ (the recurring leak is fixed); local copy retained on disk
  • (optional) a scoped rule-sweep / ownership-walk run to watch the bounded caps + unverified flow end-to-end

🤖 Generated with Claude Code

0xaslan and others added 5 commits June 23, 2026 18:24
…olidate unverified findings

Sibling half of the predict-audit harness rework whose orchestrator + SKILL.md
half already landed in main via the reference-tick admission PR. Read-only on
packages/* — audit tooling only.

ownership-walk.workflow.js, rule-sweep.workflow.js:
- Bound by construction: MAX_ROUNDS default 12->3 (ownership-walk) and 10->3
  (rule-sweep), add verifyCap (60), severity-gated verify (cleanup-tier reported
  raw, no subagent). Agents <= maxRounds*units + verifyCap, so neither can run
  away to the agent cap.
- Each now emits an `unverified` array (Info/Low/cleanup triaged out of verify).

consolidate.py:
- Collect the new `unverified` key from every harness: record it under its own
  key in findings.json (so track.py never floods the tracker with low-priority
  noise), keep it in the DROPPED-0 accounting (honest, no silent drop), and
  surface it in an Unverified report section. Re-syncs the consolidator with the
  already-merged orchestrator, which emits `unverified`.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Committed in 219a910 and left tracked, so .gitignore (/docs/, /docs/superpowers/) never applied — tracked files bypass ignore rules, which is why these specs kept reappearing in PRs. git rm --cached untracks it (local copy kept); the existing rule now keeps it out.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@0xaslan 0xaslan merged commit 177e62b into main Jun 25, 2026
3 checks passed
@0xaslan 0xaslan deleted the at/predict-audit-bound-siblings branch June 25, 2026 18:36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants