# Synthesis Skills

Proven AI agent skills for code review, content creation, project management, and more. Built on the [Agent Skills](https://agentskills.io) open standard and portable across Claude Code, OpenAI Codex, Cursor, GitHub Copilot, and other capable agents.

## What's new

**Disclosure governance by precedent, not blacklist (July 2026).** `synthesis-disclosure-policy` distinguishes the names you deliberately publish in your own biography from disclosures nobody approved, backed by an evidence-cited precedent ledger. `synthesis-git-hooks` **v2.2.0** enforces it by publication surface: your site repos get full protection minus your ledgered names, public OSS repos stay pinned strict, and a missing or unverifiable ledger fails closed. See the [4.8.1 release notes](CHANGELOG.md).

**Trustworthy resumption and safe retirement (July 2026).** Activation, handoff, and SessionStart context now detect stale project checkouts by comparing the record with its fetched upstream, and handoff verifies that both client envelope formats carry identical context. `synthesis-project-management` **v1.8.0** makes lease-managed boards self-declaring — a machine without the lease config refuses to write rather than losing changes silently, with a sanctioned `lease-disable` path — and adds `retire_worktree.py` for fail-closed, remote-verified retirement of merged worktrees. See the [4.7.0 release notes](CHANGELOG.md).

**Symmetric verification and cross-machine coordination (July 2026).** Conformance checks, doctors, and installers now resolve the Claude and Codex CLIs through overrides, `PATH`, and documented install locations, so the same verification runs from either client's shell. `synthesis-project-management` **v1.7.0** adds an opt-in git-backed coordination lease — an atomic ref compare-and-swap on a shared remote — for safe same-resource sessions across machines, and claim-overlap detection now matches mixed absolute, `~`, and relative claim spellings. See the [4.6.0 r