JetThoughts Claude Code plugins

Marketplace repo. Install:

claude plugin marketplace add /Users/pftg/dev/claude-plugins
claude plugin install j-delivery@jetthoughts
claude plugin install j-unfix@jetthoughts

j-delivery

The autonomous delivery contract (4-eyes, evidence standard, WIP=1) as a load-on-demand skill, the /deliver kickoff command, the async-first SOP, and the core author/verifier agent roster. The consuming repo's own instructions (CLAUDE.md / AGENTS.md) override the plugin on every conflict - the plugin is the default, never the authority.

Canonical origin: jetthoughts/jetthoughts.github.io ADR-0005; that repo keeps its project-specific appendices (tool snapshot, domain map, gates).

Versioning: semver git tags; consumers upgrade deliberately via claude plugin update. CHANGELOG per plugin.

j-unfix

Jurgen Appelo's unFIX pattern library for organisation design - Bases, the seven Crew types, Forums, Captains, Chiefs, Chairs, and the sixteen Role Attributes - written from unfix.com rather than from a remembered summary. Includes a section on applying it to an AI-agent organisation that separates what transfers from what degenerates, because Chiefs are defined by recruitment and compensation and agents have neither.

Use it for org structure, team topology, decision rights, reteaming, or naming the seats in an agent org.

j-harness-setup

Project harness setup: discover current capabilities, reconcile context, ask targeted questions, reuse installed skills and review changes before editing. Instruction-only — no scripts, no generator, no MCP servers. Reuse an available skill-creator instead of installing a duplicate.

claude plugin install j-harness-setup@jetthoughts --scope project

Invoke /j-harness-setup:setup with one project outcome. Installation is optional: use a reviewed checkout with claude --plugin-dir /absolute/path/to/plugins/j-harness-setup for a temporary trial. See setup and safety boundaries.

This plugin currently ships three skills: setup, plus omniroute-manager and browseros-neo, which are machine-environment skills that belong in their own plugin. See SKILLS_REVIEW.md (T2.4) — the plugin README carries the same caveat.

Adding a skill to this repo

Create the directory and the file — never a symlink:

mkdir -p plugins/<plugin>/skills/<skill>
$EDITOR plugins/<plugin>/skills/<skill>/SKILL.md

Then add or bump plugins/<plugin>/.claude-plugin/plugin.json, and make sure the plugin is listed in .claude-plugin/marketplace.json and the root plugin.json.

Do not symlink a skill into ~/.claude/skills. That was the documented recipe until 2026-09-30 and it produced 22 self-referential symlinks committed to git, a skill discovered 592 times instead of 99 by anything that followed links, and a plugin cache whose entries had empty targets — which is why several plugins reported failed to load. Delivery is the marketplace only; NAMING.md and INSTALL.md carry the reasoning.

Before committing:

python3 scripts/validate_skills.py     # 0 errors required

The same check runs in CI (.github/workflows/validate-skills.yml). Install it locally with bash scripts/install-hooks.sh. All of the defects in SKILLS_REVIEW.md were mechanically detectable, so this is what keeps them from coming back.

Vault-coupled skills stay in the vault. jt, jt-exec-ops, jt-research-brief and jt-sprint read ~/dev/pkm/.ai/state/* and the vault's notes; extracting them would leave a plugin that only works in one checkout and split the content away from the repo that owns it.