KeiSeiKit-1.0/_capabilities/verify/fork-audit/text.md at 5f7a5b2639b790c78c0fb77bd8f69208c09e23fd

Parfii-bot 5f7a5b2639 fix(wave18): 8 HIGH audit findings closed + three-role pipeline actually built

47 crates, 801 tests green (up from 771 at v0.34.0). Wave 18 audit
found 8 HIGH findings across architect/critic/security/validator. All
closed. Three-role pipeline REBUILT after validator discovered Wave 16
commit was a half-commit (files claimed but never tracked).

## A. Three-role pipeline (REBUILD — was missing from v0.33.0 despite
      CHANGELOG claim)

Files validator flagged absent: _roles/auditor.toml + merger.toml,
4 _capabilities/{policy/git-ops-scope,output/verdict,output/merge-result,
verify/fork-audit}/text.md, kei-spawn/src/{pipeline,precedent}.rs,
pipeline_smoke.rs + pipeline_unit.rs tests. ALL NOW REAL (verified by
git log --all and `ls`).

- auditor role: claude-subagent-type=critic, handoff=[merger]
- merger role: git-ops scope, claude-subagent-type=infra-implementer,
  leaf (empty handoff)
- 5 capability text.md (+ capability.toml for each) defining contracts
- kei-spawn pipeline.rs (171 LOC): pipeline_from_role, derive_steps,
  emit_pipeline_json, scaffold_downstream_tasks
- kei-spawn precedent.rs (118 LOC): env-gated advisory shell-out
- --pipeline flag on spawn subcommand
- +11 tests (pipeline_smoke + pipeline_unit)

## B. kei-fork — 4 HIGH fixes (Critic F1+F7a, Security #3+#4)

- `git add -A` → explicit path list from ls-untracked + ls-modified,
  with exclusion filter for .DONE / .KEI_FORK_META.toml / _archive/ /
  _forks/. No more merge bleed. +1 regression test.
- create() rollback: on write_meta or ledger_fork failure, worktree
  + branch cleaned. +1 test via KEI_FORK_FORCE_LEDGER_FAIL=1.
- worktree_add arg injection: added `--` sentinel + is_safe_refname()
  validator (refuses dash-leading, NUL, ..). +3 tests.
- PATH hijack: KEI_FORK_GIT_BIN env override for all Command::new(git).
  +1 test.

## C. kei-spawn — 2 HIGH fixes (Security #1+#2)

- HTTP body unbounded DoS: MAX_BODY_BYTES=10MiB + content-length
  pre-check + streamed cap (io::Read::take) for chunked encoding.
  +2 feature-gated tests.
- PATH hijack: KEI_LEDGER_BIN env override already existed at
  ledger_sh.rs:15; documented precedence + added 4 regression tests
  locking the 3-tier lookup order.

## D. kei-ledger-sign — 1 HIGH fix (Security #2)

- save_keypair atomic POSIX open(2) O_CREAT|O_EXCL + mode 0o600 +
  rename(2) into place. No race window where key is world-readable.
  +2 tests.

## E. spawn_from_task rollback (Critic F7b)

- register_in_ledger helper: on ledger fork failure, rollback_task_dir
  before error propagation. +1 test spawn_rolls_back_task_dir_on_ledger_fail.

## Audit summary

- architect: GO conditional (taxonomy 19% — defer)
- critic: HIGH closed, MEDIUM debt logged
- security: 4 HIGH closed; MEDIUM (tar symlink, watcher symlink) tracked
- validator: CHANGELOG no longer lies — three-role pipeline is real
- patent-compliance: GO / LOW risk unchanged

All 8 HIGH blockers from Wave 18 consolidated audit → GREEN.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-04-23 20:54:59 +08:00

2 KiB

Raw Blame History

Fork audit — 6-point checklist

When reviewing a writer's fork diff, your return MUST address each of the six points below. Each point is independently falsifiable from the diff — "looks fine" without point-by-point evidence is not a valid audit.

Diff coverage. Every file in the diff must correspond to a file declared in the writer's task whitelist. Orphan writes (outside whitelist) → FAIL. Include the exact path of any orphan in your verdict.
Test evidence. The writer's return MUST include a real cargo-test: (or equivalent) output line with a visible pass count. "Tested mentally" / "tests should pass" / any paraphrase → FAIL. Cross-check the test count matches new test files in the diff.
Scope adherence. No edits outside the writer's declared whitelist. Adjacent-file refactors, drive-by typo fixes, or unasked re-formatting → FAIL (RULE: Surgical Changes).
Capability enforcement. If the writer's role required capabilities (e.g. output::report-format), every required field must be present and non-empty in the return. Missing field → FAIL.
Constructor-pattern LOC limits. Any new .rs file must be ≤200 LOC; any function ≤30 LOC. Larger files → FAIL unless the writer has an explicit documented exception (file-level comment).
Blocker disclosure. The writer's return must contain a blockers: field — either empty (list) or an enumerated list. Silent dropping of known issues → FAIL. Silence = FAIL, not PASS.

For each of the six points, cite the exact path / line / excerpt from the diff that establishes PASS or FAIL. The verdict is derived from these six points:

PASS — all 6 points evidence PASS.
FAIL — any point evidence FAIL. Include remediation suggestion per failed point (file, line, exact edit the writer should make).
INCONCLUSIVE — point N cannot be evaluated from the available diff (e.g. tests didn't run, CI output missing). State which point and what would make it evaluable.

2 KiB Raw Blame History

Fork audit — 6-point checklist

2 KiB

Raw Blame History