One loop
Align on the goal, fix an anchor (the exact check that observes the result), act in small increments, verify against the anchor. Every entry skill runs it.
anchorloop
Agent workflow skills and loop extensions for pi. Every task enters through its own skill, picks a playbook, and runs one shared loop, Align → Anchor → Act → Verify, that the extensions watch.
An agent that is told to “just fix it” skips the part that makes a fix believable: say what the user wants, name the check that will observe the result, change the code in small steps, and report what the check saw. anchorloop is a set of pi skills and extensions that make that the default way every task runs.
An entry skill per task
coding, investigating, inspecting, answering, writing, visual-designing, integrating, deploying, and solid-loop, each with playbooks and its own anchor.
A book that watches
A loop book reads edits, check results, reviews, and landings as evidence, so the phase of a loop comes from what happened, not from what the agent says.
Merge trains
Reviewed branches enter a shared branch as one train: one assembly, one review, one run of the heavy gates, then deploy target by target.
Friction becomes rules
An instruction that misled a run is recorded, classified, and fixed at its root by solid-loop, so the same detour does not return.
Handoffs that keep evidence
A session that leaves a loop hands over its book’s id; the next session resumes with what each unit already proved.
Read next
Section titled “Read next”InstallOne command, plus pi-better-show for review pages.
Quick startFour steps through a first task.
ConceptsThe loop, the four kinds, anchors, units, landing, and friction.
SkillsThe skill tree and what each entry skill anchors on.