pipe-skills

Tracker

• Track work in GitHub Issues, on the GitHub project Pipe's skills of felipeam86.

Statechart

• The statechart in docs/statecharts and the skills describe one behavior, so they change in the same commit, in both directions: a skill change that adds, removes, or moves a step, a state, or a rule changes Lifecycle.tla and view.mjs, and a model change changes the skill text it describes. Run TLC with the command for each chart in docs/statecharts/README.md, and paste both new charts into README.md, because a chart that drifts from the skills tells the user a wrong story.
• When you put a new chart.mmd into README.md, remove its Legend block and its class lg lines, and add one classDef line for each phase color that the chart uses, with the colors of the chart in README.md. The command always writes the Legend, which Mermaid cannot place at a fixed position, so README.md gives the colors in one line of text, and the command writes no classDef for the phase colors.
• Each time you change a skill, read the model, and say in the commit message whether each chart changed, so that a skill change that skips the model shows in the history.

Skills

• A skill gives the model clear, short instructions on how to behave and why, and leaves the method to the model, because a model follows short, clear instructions well, and each step that the skill prescribes is more text to maintain and to review. The cleanup of ship in #8 is the case.
• A skill reads the values of a project, such as its Linear team and project, from the CLAUDE.md of that project and never holds them, because the plugin serves every project that installs it.
• A skill that merges, closes an issue, or deletes a branch sets disable-model-invocation: true, so that only the user starts it. ship and autonomous-loop are the case.
• A skill names another skill as pipe-skills:<name>, and a subagent prompt reads a skill at ${CLAUDE_PLUGIN_ROOT}/skills/<name>/SKILL.md, so that the reference holds wherever the plugin is installed.
• A skill owns its behavior. The README.md gives each skill one row with one example, plus the chart, so a new, renamed, or removed skill changes its row in the same commit.
• A guard in a skill never stops autonomous-loop: when the guard asks the user for a go, the loop continues without it, because the loop exists to work through a queue with no stop for the user.