gaia-research/gaia-skill-tree / .agents/skills/trust-appraise
trust-appraise
VerifiedDry-run Trust Magnitude for proposed named skills or suites before curation. Use when deciding whether a proposed suite deserves A/S treatment, when checking Fusion Score structural inputs, when comparing repo stars vs suite component counts, or when asked to appraise a candidate before adding it to the registry.
/trust-appraise
Runs a non-mutating Trust Magnitude dry run. Formulas, dedup, inheritance, grade
floors and star gates are not restated here. The canonical source is
docs/agents/curation-authority-map.md,
which is checked against src/gaia_cli/trustMagnitude.py by
tests/test_curation_authority_map.py.
Two modes:
- Registry mode (
--skill) appraises an already-curated skill. Acontributor/skillid readsregistry/named/<contributor>/<skill>.mdand resolves itsgenericSkillRefagainstregistry/nodes/. A bare id reads the generic node. Use it during L4 human review to see the real TM and per-row artifact scores before signing off. - Proposal mode (
--repo) appraises a proposed suite from live GitHub signals before curation. It is the pre-curation companion to/trust-appraise-all.
Usage
# Appraise one or more curated skills (use during L4 human review)
PYTHONPATH=src python3 scripts/trust_appraise.py \
--skill rico-favor/implement-with-discernment \
--skill caioribeiroclw-pixel/evidence-attestation
# Appraise a proposed suite (pre-curation)
PYTHONPATH=src python3 scripts/trust_appraise.py \
--repo gsd-build/get-shit-done \
--components 5 \
--evidence-path docs/INVENTORY.md
# Default suite appraisal set
PYTHONPATH=src python3 scripts/trust_appraise.py
# Machine-readable output
PYTHONPATH=src python3 scripts/trust_appraise.py --skill foo/bar --json
Frozen skills: a frozen skill's standing is its locked TM/grade (gaia_cli.frozen), never
the recomputation. The text output prints FROZEN - locked TM ... / ...; current recomputation ... / ... and the JSON carries frozen, recomputedTm, recomputedGrade. Report the lock as
the standing and the recomputation only as information; the rows below it are the recomputation.
Scope labels: always report them
Every result carries scope, genericResolution and totalStatus (authority map §2):
| Mode | scope | genericResolution | totalStatus |
|---|---|---|---|
--repo proposal | named-own/proposal | not evaluated | partial |
--skill named, generic ref resolves | effective named | resolved | final |
--skill named, generic ref missing from the registry | named-own | not evaluated | partial |
--skill named with no genericSkillRef | effective named | not applicable | final |
--skill generic node | generic | not applicable | final |
A partial total is never the effective named TM. Say "generic: not evaluated", not
"inheritance disallowed". Generic-layer evidence that a skill later inherits still counts,
even when it does not prove implementation usage.
Proposal-mode metrics and their status
The proposal builds a temporary skill from its own rows only:
github-stars-ownat the evidence blob URLrepo-ownat the repository root- a structural
fusion-reciperow, which scores 0 TM and feeds only the informational Fusion Score
skillCountInRepo is recorded and never divides a score. Each collected metric is reported
in metricClaims with a verified / reported / inferred / unknown status and its
repro command (authority map §5):
contributorContributionSumis the sum of GitHub per-contributorcontributions. It is not a count of unique commits. The engine'srepo-owninput field is namedcommits, so the proposal fills that input from the sum and labels itinferred. Never report it as "N commits".- Repository metrics describe the whole repository. Skill-file installability (a verified
links.githubblob/URL) is a separate check.
Important caveat
A high dry-run score is not promotion approval. Use this skill to surface evidence
quality and TM bias before L4 human review, never to bypass review. arxiv rows with 0
citations score 0 TM whatever their trustNumber. social-signal rows admit both eligible first-party and third-party demonstrations of the target skill (sameAsAuthor = 1.0, publishCooldownDays = 0 — no 30-day hold, views >= 1000 floor, (type, url) dedup, 3-row plateau, 80 TM cap; unrelated uploads do not qualify), need verifiable view counts (fabricated numbers inflate scores directly), and must record provenance (creatorRelation, relevanceNote). proxy-containment strictly requires a qualifying external, non-first-party repository (externalStars >= 10000), and synthetic self-attestation rows are never added to bypass the 49.9 TM suite-component shared-repository cap (#2473). 4★ strictly requires grade A (TM ≥ 100), and grade S (5★) needs an independent witness. Both rules are in authority map §3.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |