gaia-research/gaia-skill-tree / .agents/skills/dev-calibrate
dev-calibrate
VerifiedPlaybook for correcting a named skill's star level to match its computed Trust Magnitude grade, with an independent post-calibration verification pass and a CLI-logged timeline event (never hand-written). Use when: recalibrating a skill's level, fixing a stale or drifted star rating, running a batch of `gaia dev calibrate` corrections, or confirming a past calibration actually landed. Triggers: "calibrate <skill>", "recalibrate the trust magnitude backlog", "promote/demote <skill> to N stars", "why is <skill> still N stars", "verify a calibration actually landed", "Hall of Heroes ranking is wrong".
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.1 KB |
SKILL.md
View on GitHub---
name: dev-calibrate
description: >-
Playbook for correcting a named skill's star level to match its computed Trust
Magnitude grade, with an independent post-calibration verification pass and a
CLI-logged timeline event (never hand-written). Use when: recalibrating a
skill's level, fixing a stale or drifted star rating, running a batch of
`gaia dev calibrate` corrections, or confirming a past calibration actually
landed. Triggers: "calibrate <skill>", "recalibrate the trust magnitude
backlog", "promote/demote <skill> to N stars", "why is <skill> still N stars",
"verify a calibration actually landed", "Hall of Heroes ranking is wrong".
playbookVersion: 1
class: B
objective: >-
Take one named skill whose stored star level has drifted from its computed
Trust Magnitude grade, and land it at the correct level with pre-flight,
decision, and independent post-calibration verification all provable — not
merely asserted by the calibrate command's own stdout.
capability: >-
Reading a Trust Magnitude leaderboard row against a skill's current stored
star level, deciding whether the gap is a real promotion or demotion versus a
case the CLI's pre-flight should legitimately block (missing verified
links.github, zero evidence), applying the correction through the canonical
CLI verb, and re-inspecting the same skill afterward to confirm the mutation
actually took.
preconditions:
- the target skill_id exists as a named skill under registry/named/<contributor>/<name>.md
- gaia whoami resolves to verifier, override, or bootstrap, not denied
- the working tree is on a review/meta/* or dev/* branch, never main
- the active gaia is this repo's source, not a stale pipx install (compare `python3 -m gaia_cli.main --version` to the version in pyproject.toml)
steps:
- id: snapshot-target
run: python3 scripts/inspectTrustMagnitude.py --skill {skill_id}
proves: current stored level, computed trustMagnitude, and overallTrustGrade are captured for skill_id
- id: decide
judgment: PROMOTE | DEMOTE | NO_CHANGE | BLOCKED_MISSING_EVIDENCE
rules: >-
Target level is the G7 star mapping of overallTrustGrade (S=5★, A=4★,
B=3★, C=2★, ungraded=1★), held at the skill's historical rank floor
(§10.10) if it previously reached 4★+. NO_CHANGE when current level already
matches the mapping. BLOCKED_MISSING_EVIDENCE when the skill has zero
evidence rows and zero suiteComponents driving its Trust Magnitude — the
grade is not a real signal to act on until evidence is sourced.
- id: calibrate
run: gaia dev calibrate {skill_id} {target_level} --no-build
proves: level frontmatter updated and a rank_up or demote timeline event appended automatically by the CLI — never hand-write this event
- id: verify-landed
run: python3 scripts/inspectTrustMagnitude.py --skill {skill_id}
proves: re-inspected stored level equals decide's target_level — confirmed against live state, not trusted from calibrate's own stdout
- id: relabel-dependent-origins
judgment: RELABEL_APEX_ORIGINS | NO_DEPENDENT_ORIGINS
rules: >-
Find every named skill whose fusion-recipe.origins or suiteComponents
includes skill_id. Re-run origin-grade resolution and re-inspect any Apex
Gate predicate state that depends on graded-origin counts. Land any
resulting calibration change through gaia dev calibrate (never hand-edit
an origin label or Apex predicate).
- id: rebuild-docs
run: gaia dev docs
proves: Class S artifacts (docs/graph/*, docs/api/v1/*) regenerated to reflect the new level
- id: verify-docs-clean
run: python3 scripts/build_docs.py --check
proves: exit 0 — the committed artifact set has no undeclared drift beyond the documented warn-only categories
stopConditions:
- decide resolves BLOCKED_MISSING_EVIDENCE — land in the review packet for sourced-evidence follow-up, do not force a level change
- the calibrate step's pre-flight rejects the target (missing verified links.github blob for a 3★+ promotion, or a 4★+ target whose Trust Magnitude / Unique-vs-Suite branch gate fails — lowering a level is never gated) — land in the review packet, the fix is evidence work, not a retry
- verify-landed shows the stored level still does not equal target_level after calibrate ran clean — escalate to the founder queue, do not re-run blindly
proof:
- snapshot-target and verify-landed both ran, and verify-landed's stored level equals decide's target_level
- the skill's frontmatter timeline carries a rank_up or demote event for this change, logged by the CLI, never hand-written
- build_docs.py --check exits 0 after rebuild-docs
done: >-
skill_id's stored level matches its computed Trust Magnitude grade under the
G7 mapping, the change is provable from a re-inspection run after calibrate
(not from calibrate's own stdout), the timeline carries the CLI-logged event,
and Class S docs are regenerated and clean.
---
## Overview
This playbook corrects a named skill's stored star level when it has drifted from
its computed Trust Magnitude grade. The objective is not just to run
`gaia dev calibrate` — it is to make the correction *provable*: a pre-flight
snapshot, an explicit promote/demote/no-change/blocked decision against the G7
mapping, the CLI-driven mutation itself (which alone appends the timeline event —
never hand-write it), and an independent re-inspection afterward that confirms the
stored level actually changed, rather than trusting the calibrate command's own
stdout.
## Batch use
For a backlog of N skills, repeat `snapshot-target` through `verify-landed` once
per `skill_id`. `rebuild-docs` and `verify-docs-clean` are expensive registry-wide
rebuilds — run them once at the end of the batch, not after every individual
skill. This is an efficiency note only; it does not change the per-skill spine or
its proof obligations.
## Reference
- `founder/steward/PLAYBOOKS.md` — the playbook contract this file opts into.
- `python3 scripts/inspectTrustMagnitude.py --leaderboard` — finds calibration
candidates in bulk by ranking all named skills by Trust Magnitude.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |