無料GitHub で公開

gaia-ingest-batch

Batch wrapper for /gaia-ingest. Ingests a bounded set of already-verified evidence rows through CLI-only writes, uses --no-build on every row, appraises Trust Magnitude for every affected named skill, then runs exactly one build and validation pass. Use for an L4-approved intake after /ev-pipeline; benchmark-result rows also require a Phase 2B report and explicit human gate approval.

gaia-research/gaia-skill-tree

インストール方法を見る

含まれるファイル（1）

SKILL.md3.9 KB

SKILL.md（原文）

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Gaia Ingest Batch

This is a coordinator, not a second evidence-ingestion implementation. Every row follows the contract in /gaia-ingest; this skill only sequences verified rows efficiently and finalizes the resulting registry artifacts.

Input manifest

Prepare a reviewed manifest with one row per source:

rows:
  - skill: firecrawl/web-scrape-integration
    source: https://www.youtube.com/watch?v=...
    evidenceType: social-signal
    payload:
      views: 8510
    sourceStartedAt: 2025-07-20
    notes: "Third-party tutorial explicitly demonstrates page scraping."

The manifest must state a source URL, Evidence Type, source-start date, verifiable numeric payload, factual notes, and attribution scope for every row. Exclude any deferred candidate. Never infer an Evidence Type or a metric from a summary. For benchmark-result, attach the Phase 2B benchmark-source report and human approval for the lane: verified rows need CI/verifier attestation plus runAt, datasetHash, and benchmarkInputHash; reported rows may be public claims or mirrored benchmark evidence approved by the human gate and do not need those reproducibility fields. rejected, pending, candidate, retired, or unknown benchmark sources/rows must not enter ingestion as scoring rows. A catalog status: rejected is the blacklist. appliesToGenericSkillRefs is catalog metadata only; it does not create a named score without a named benchmark-result row.

Procedure

Verify /ev-pipeline completed for the manifest’s sources and link health is recorded. Reject dead, unverified, duplicate, or scope-mismatched rows. For benchmark-result, require Phase 2B + Phase 3 + Phase 4 completion and explicit human gate approval before any row is ingested.
For each row, invoke the /gaia-ingest contract and execute its exact gaia dev evidence ... --no-build command. Process one row at a time and stop on the first CLI preflight or source-verification failure.

Appraise each affected skill after its final row:

PYTHONPATH=src python3 scripts/trust_appraise.py --skill <contributor/skill-id>

Present proposed calibrations. Do not calibrate without explicit operator approval. If approval is already recorded, run each approved calibration with --no-build.
Hand off the branch-close to /gaia-review-meta-close. Do not run the build/validate/stage/PR steps here — that skill owns the single build, the calibration+Origin gate, suite wiring via gaia dev fuse (so suiteComponents survives the build), upstream-naming correction, the LF-renormalized artifact allowlist (dropping CRLF churn and blocking leaks), the UTF-8-safe validate, and the PR. Pass it the branch, the affected contributor handles (for badge/og staging), the per-skill appraised TM/grade from step 3, and the intake issues to Resolves. This skill stops at "evidence ingested + appraised."

Output

Report, for every row: CLI command, source verdict, Evidence Type, row grade, TM contribution, and duplicate/scope decision. Report, for every skill: final TM, Overall Trust Grade, current level, and any calibration proposal. Then hand these facts to /gaia-review-meta-close for the gated close-out.

Route suite creation only after components are ingested and appraised — the capstone/suiteComponents wiring happens inside /gaia-review-meta-close (gaia dev fuse), or via /gaia-fuse-full-suite for a standalone suite build.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

ログインしてレビューを書く

同じリポジトリのスキル

概要と使いどころ

ci-churn

無料

Measure CI churn cost for a pull request: how many commits and how many CI-compute seconds were burned on avoidable "fix the CI" push rounds. Use when you want to understand the efficiency cost of a PR's iteration cycle, get actionable pre-push check suggestions, or include a churn summary in a pipeline report. Triggers: "ci churn", "how much CI did we burn", "avoidable commits", "ci efficiency", "wasted CI time", "churn report", "/ci-churn". Called automatically by /fp-drift at pipeline close.

日本語の概要は準備中です。原文の説明を表示しています。