Changelog

• Agent Safety Protocol now also requires a test for every change, a definition of done (compiles, lints, tests pass, no secrets, only targeted files), no debug leftovers, and a one-line provenance note on commits and PRs.
• The generated .agent/knowledge.md can now arrive pre-filled from a scan of your repo (commands, layout, conventions) when generated by Sync or the CLI, and Sync never overwrites it once your agents start maintaining it.
• New optional "Learned from Code Review" section: recurring points your maintainers make on pull requests become rules, proposed in a Sync PR for review and always lower priority than the security rules.
• Generated files no longer carry runs of blank lines (fewer tokens, cleaner diffs).
• New: Agent-Readiness Score (/score) with a README badge and JSON API, and an offline CLI (agent-hub score / init) for local use and CI gating.

• New role-aware Security Checklist: injection, XSS, CSRF, object-level authorization, rate limiting, JWT/password/MFA, CORS, uploads, webhook signatures, SSRF, row-level security, secrets and logs — only the groups relevant to your selected stacks.
• New Agent Safety Protocol: scope limits, package verification, untrusted-input isolation, secret scanning, a 5-attempt circuit breaker, and an explicit human-approval list.
• New Project Knowledge Base: agents read and update a compact .agent/knowledge.md each task instead of re-scanning the repo, cutting token use; a seed file is included in every bundle.

Site-only change — doesn't affect files you've already generated.

• New: Agent Hub Sync, an optional GitHub App that watches your repo and opens a pull request when your dependencies change or these guardrails are updated, instead of relying on you noticing the banner on this site.
• Runs on every push (re-detects your stack) and once a day (checks for a new content version) — see /sync for setup.
• Site-only — this doesn't change what a manual download from this page generates, so it doesn't trigger the out-of-date banner below.

Site-only change — doesn't affect files you've already generated.

• Added a password-protected /admin dashboard for the site owner: total visits, visits per day, most-picked stacks and tools, generation-mode split, and CI-guardrail adoption rate.
• Purely operational — this release changes nothing in the files you download, so it doesn't trigger the out-of-date banner below.

• Custom Team-Specific Directives: your own free-text rules, layered into the generated document with explicit precedence over the generic guidance.
• Optional CI guardrail check: a GitHub Actions workflow + script that fails a PR on committed secrets, disabled lint/type rules, or a force-push/history-rewrite command — so the rules are enforced, not just suggested.
• Shareable config links: your exact stack + mode + target + custom-rules selection encoded into one URL a teammate can open to get the identical setup.
• Repo-aware auto-detection: paste a package.json/requirements.txt/go.mod/etc. and the matching stacks are pre-selected for you.

• Operational Guardrails & Safety Boundaries: explicit Always-allowed / Ask-first / Never-allowed tiers for what an agent can do unsupervised.
• Verification Workflow: a mandatory self-verification loop plus a real per-stack lint/type-check/test/build command table, instead of a vague "make sure it works."
• Git & Pull Request Conventions: branch naming, Conventional Commits, and a PR-description checklist, appearing once per document.

• 22 stacks across six categories, combinable into one cross-stack-aware agent or split into per-stack files.
• 10 Core Operating Directives, Universal Guardrails, and mandatory end-of-session Task Summary & Impact reporting.
• Output packaged for 8 AI coding tools: GitHub Copilot, Claude Code, Cursor, Windsurf, Cline, Continue.dev, Aider, and AGENTS.md.