How anchorloop's extensions work

The full behavior behind the README. The skills tell an agent how to work; the six pi extensions below watch the work and say what is missing, so find the section you need and skip the rest.

anchorloop’s extensions keep one loop book per open loop, read what the loop has done from evidence instead of self-report, and turn that into a phase, a status line, a warning, or one more turn.

flowchart TD
  agent["Agent<br/>todo and<br/>anchorloop_stage"] -->|"declares"| book
  tools["Tool results<br/>edits, bash exit codes,<br/>reviews, spec files,<br/>subagent results"] -->|"evidence"| book
  git["git refs"] -->|"landings"| book
  book[("Loop book<br/>one YAML file per loop")] --> phases["Phase machine<br/>current, next, missing"]
  phases --> widget["Status widget<br/>below the editor"]
  phases --> gate["Anchor gate<br/>one warning line on an edit"]
  phases --> second["Second thought<br/>Jev judges the reading"]
  second -->|"next: ... or one more turn"| agent

Extension	What it does	Section
pi/book.ts	Keeps the loop book, reads evidence into it, and shows the loop’s phase.	The loop book, Evidence, Phases, Reach, Anchor runs, Subagents and units, Landing, Handoff
pi/anchor-gate.ts	Warns when an edit starts the act phase before an earlier phase is complete.	The anchor gate
pi/decision-tool.ts	The anchorloop_decision tool: a decision trail for runs the user reviews later.	The decision trail
pi/second-thought.ts	Asks Jev to check the phase machine’s reading when a run settles.	Second thought
pi/validate-tool-use.ts	Validates skills, instruction files, and specs after each edit.	Validators
pi/ttsr.ts	Time Traveling Stream Rules: a rule fires when the agent’s own output matches it.	Stream rules