using-cairn

What cairn is

Cairn is a verification substrate for AI-coordinated software development. See PLAN.md §"What this is / is not" for the full substrate summary. This skill teaches when to invoke cairn's three wrappers, how YAML specs derive silently from prose, and which spoke to load for specific tasks.

When this skill applies

Three scenarios trigger this skill:

1. YAML lifecycle — Is there a specs/ dir with cairn YAML? YAML derives from prose. Read spoke yaml-authoring.md.
2. Wrap routing — Is the agent executing an implementation plan inside a cairn-tracked claim? Route to the appropriate wrap.
3. Code reviewer dispatch — Is a reviewer agent being dispatched against a rubric gate? Read code-reviewer-pattern.md.

Wrap routing rules

Situation	Use
Executing a plan via subagent dispatch inside a cairn-tracked repo	cairn:subagent-driven-development-with-verdicts
Executing a plan outside a cairn-tracked repo (no specs/)	superpowers:subagent-driven-development
Verification before claiming complete, while holding an active cairn claim	cairn:verdict-backed-verification
Verification before claiming complete, no active cairn claim	superpowers:verification-before-completion
Brainstorming, plan writing, test-driven development, receiving code review	superpowers:* originals unchanged. No cairn wrap.

YAML lifecycle

In a cairn-tracked repo, specs/ YAML is always derived from prose. The human never edits YAML directly. Requirements YAML derives from docs/superpowers/specs/*.md (brainstorming output); task YAML derives from docs/superpowers/plans/*.md (writing-plans output). Derivation is deterministic and byte-identical on re-run. See yaml-authoring.md for the full protocol.

Hash placeholders banner

When binding a verdict, producer_hash and inputs_hash use provisional Ship 3 placeholders.

These hashes are placeholders. They do not reflect toolchain version or input state. Verdicts bound with these values are NOT safe to rely on for cross-run drift detection.

See hash-placeholders.md for the recipe and the future-replacement plan.

Invocation rule

This skill MUST be invoked explicitly by the orchestrating session after each of superpowers:brainstorming and superpowers:writing-plans commits. Agent-noticing is unreliable. For harness-enforced belt-and-suspenders drift detection, enable the cairn Stop hook via cairn hook enable. The hook supplements skill discipline; it does not replace it, and it only catches source-hash drift (not skipped verdicts or orphan claims — those are guaranteed by cairn's mutation path itself).

Routing to spokes

Task you're about to do	Load first
Author or regenerate YAML from prose	yaml-authoring.md
Compute producer_hash or inputs_hash for a verdict	hash-placeholders.md
Read or write the # cairn-derived: comment	source-hash-format.md
Dispatch superpowers:code-reviewer against a rubric gate	code-reviewer-pattern.md

Red flags

Thought	Reality
"I'll just edit the YAML directly"	YAML is derived. Edit the prose; regeneration follows.
"The prose spec is fine, skip the elicitation"	Elicitation checks for cairn-required fields not present in prose. Skipping = malformed YAML downstream.
"Verdict is close enough without evidence"	Core Invariant 3 — no verdict without hash-verified evidence.
"Agent said it's done, skip cairn task complete"	Core Invariant 10 — cairn events since must show the completion.