Skip to content

Fix workflow contradictions and add run tooling (0.11.0) - #3

Merged
James-Chapman merged 1 commit into
mainfrom
review-fixes-0.11.0
Nov 11, 2026
Merged

James-Chapman merged 1 commit into
mainfrom
review-fixes-0.11.0

Conversation

@James-Chapman

Copy link
Copy Markdown
Contributor

Result of a full review of the skill. The evidence bar and the execution-consent model are unchanged.

Fixed:

  • Output directory: give a resolution order, because "./security-audit" is inside the target when the agent starts there.
  • Carry open findings forward into each new run and add a "fixed" status, so a later report no longer drops findings nobody fixed.
  • Use the hardened Git form in the phase commands, not only in execution.md.
  • Map each reference-file section to the phase that reads it, and move per-domain recon detail to a new domain-notes.md (one line changed in each of the 33 domain files).
  • Tool rules: toolchain launchers that download or run what the target names, target ignore files, gitleaks --redact, and an unprivileged container user.
  • Merge safety rules 6 and 9 (patches); rule 9 now covers urgent discoveries.
  • Define the memory-safety, race and parser-differential claim types.
  • Verify chain components before chains; handle a rejected link.
  • Remove references to a script and a CWE catalogue that did not exist.

Added:

  • assets/plan-template.md, scoped Full audits, root-cause-at and first-seen in the finding template, a checkpoint after recon with owner-attested facts, a countable budget, orphan reset on resume.
  • Windows host guards and a Windows Sandbox recipe.
  • scripts/audit_run.py (check, stats, export to JSON and SARIF) with unit tests; the validator now checks the reference files' structure.
  • Five evaluation cases, trigger queries and a workspace-check tool.
  • CI workflow, .gitignore, CHANGELOG.md, license in the frontmatter.

Not verified: the new evaluation cases have not been run against an agent, and the container flags and the Windows Sandbox recipe are written from documentation. See CHANGELOG.md.

Result of a full review of the skill. The evidence bar and the
execution-consent model are unchanged.

Fixed:
- Output directory: give a resolution order, because "./security-audit"
  is inside the target when the agent starts there.
- Carry open findings forward into each new run and add a "fixed"
  status, so a later report no longer drops findings nobody fixed.
- Use the hardened Git form in the phase commands, not only in
  execution.md.
- Map each reference-file section to the phase that reads it, and move
  per-domain recon detail to a new domain-notes.md (one line changed in
  each of the 33 domain files).
- Tool rules: toolchain launchers that download or run what the target
  names, target ignore files, gitleaks --redact, and an unprivileged
  container user.
- Merge safety rules 6 and 9 (patches); rule 9 now covers urgent
  discoveries.
- Define the memory-safety, race and parser-differential claim types.
- Verify chain components before chains; handle a rejected link.
- Remove references to a script and a CWE catalogue that did not exist.

Added:
- assets/plan-template.md, scoped Full audits, root-cause-at and
  first-seen in the finding template, a checkpoint after recon with
  owner-attested facts, a countable budget, orphan reset on resume.
- Windows host guards and a Windows Sandbox recipe.
- scripts/audit_run.py (check, stats, export to JSON and SARIF) with
  unit tests; the validator now checks the reference files' structure.
- Five evaluation cases, trigger queries and a workspace-check tool.
- CI workflow, .gitignore, CHANGELOG.md, license in the frontmatter.

Not verified: the new evaluation cases have not been run against an
agent, and the container flags and the Windows Sandbox recipe are
written from documentation. See CHANGELOG.md.

Co-Authored-By: Claude Mythos 5.2 <noreply@anthropic.com>
@coderabbitai

coderabbitai Bot commented Nov 11, 2026 •

Copy link
Copy Markdown

Explain Change →

Caution

Review failed

The pull request is closed.

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

@sonarqubecloud

Copy link
Copy Markdown

Quality Gate Failed Quality Gate failed

Failed conditions
D Security Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE

@James-Chapman
James-Chapman merged commit 196f037 into main Nov 11, 2026
2 of 3 checks passed

def etag_for(body: bytes) -> str:
# Scanner alert 1: "weak hash MD5".
return hashlib.md5(body).hexdigest()

if args.command == "snapshot":
entries = inventory(args.directory)
Path(args.inventory).write_text(json.dumps(entries, indent=2, sort_keys=True) + "\n", encoding="utf-8")
print(f"Recorded {len(entries)} entries")
return 0
if args.command == "compare":
before = json.loads(Path(args.inventory).read_text(encoding="utf-8"))
Comment thread scripts/audit_run.py


def file_stats(run, list_path, limit):
wanted = [line.strip().replace("\\", "/") for line in Path(list_path).read_text(encoding="utf-8-sig").splitlines()]
with:
python-version: ${{ matrix.python-version }}
- name: Install PyYAML
run: python -m pip install pyyaml==6.0.2
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