feat(hardening): add readiness-review skill - #80
Merged
Merged
Conversation
One read-only report answers two launch questions for any repo: is the code healthy, and can a user do the core jobs. Product specifics live in a per-repo .readiness-review.yaml so the skill carries nothing product-specific. Read-only is enforced in code, not prose: an allowlist refuses any non-read gh or git call, the database session must confirm read-only before a catalog-only query runs, the reviewer reads a copy without .git or .env files, and findings are rendered for a person to file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XjaMDDryQsA3hTW7QbEgDP
The reviewed repo's config is untrusted, so it must not choose where the report is written, which files are read, or which database is reached. The operator now picks the report folder and any shell variable holding a connection string; config paths that leave the repo are ignored and listed in the report. The gh allowlist also refused graphql only as the first argument; it now checks every argument. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XjaMDDryQsA3hTW7QbEgDP
A reviewed repo's config could name another repository and steer the read-only GitHub calls at anything the operator's token can read. The repo now comes from the checkout's own GitHub remote; only the operator's --github-repo overrides it. Config-supplied regexes are capped at 200 characters and skipped when they fail to compile or nest quantifiers, so a hostile pattern cannot hang the scan. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XjaMDDryQsA3hTW7QbEgDP
The reviewer's copy of the repo removed only .git and .env files, so credential files such as .npmrc, private keys, and cloud service-account JSON stayed readable. They are now excluded by one list, checked per path segment, with redaction kept as the backstop. Binary files and files over a size cap (1 MB, set only by the operator) are skipped too, and the copy reports how many files and bytes it left out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XjaMDDryQsA3hTW7QbEgDP
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds
readiness-review, a sixth skill in the hardening pack: one read-only launch-readiness verdict for any repo and product. It combines a code-health review (security, architecture, live database structure, unfinished features) with a walk through the running product (can a user do the core jobs, is it hard to use, does it look finished) and produces one report ending in a ready-to-file findings list.Safety rules (each enforced in code and tested)
gh_facts.py).db_facts.py)..git,.env*, credential files (.npmrc, keys, cloud/SSH configs, service-account JSON), symlinks, binaries and files over 1 MB, then re-checks the copy (make_review_copy.py). All repo content is passed to the AI as untrusted data.write_report.py).--exerciseis opt-in per run and needs a named test account and workspace, a spend cap and an action cap, and logs every action. Code and database stay read-only in every mode.Tests
scripts/tests/test_readiness.py), all passing: redaction, the database read-only guard, the gh allowlist, never-overwrite, the review copy's exclusions, config values that are ignored, and the scanners on sample Next.js and FastAPI repos.npm run checkandnpm run lintpass (manifests, all 62 skills load, links, README counts, private-term check, spelling).CREATE TABLE.Also updates the hardening README, both plugin manifests, both marketplace manifests, the root README and site (61 → 62 skills),
package.json, andskills.sh.json. The npm package version bump is left for the usual release PR.🤖 Generated with Claude Code
https://claude.ai/code/session_01XjaMDDryQsA3hTW7QbEgDP