AgentSkillsCN

qa-cartography

当您在`signal_experiments/`中构建/查询全仓库范围的取证索引（取证报告、SQLite FTS、经QA认证的数据存储与视图），并保持`Documents/RESULTS_REGISTRY.md`中的证据链接可搜索时使用此功能。

SKILL.md
--- frontmatter
name: qa-cartography
description: "Use when working in `signal_experiments/` to build/query repo-wide forensic indices (forensics report, SQLite FTS, QA-certified datastore+views) and to keep `Documents/RESULTS_REGISTRY.md` evidence-linked and searchable."
metadata:
  short-description: "QA cartography & search indices"

QA Cartography

Cartography is mapping, not cleanup: preserve all trace and make it navigable.

Quick start (safe + fast)

From repo root:

  1. •Refresh forensics (writes trace-like outputs under _forensics/):
  • •python tools/project_forensics.py --control-cutoff-utc-date 2026-01-10
    • •Anything with mtime before 2026-01-10 UTC is treated as legacy and must be re-vetted.
  1. •Optional: cartography integrity gate (compare latest two forensics runs):
  • •python tools/qa_cartography_gate.py
  1. •Build/refresh fast local search (SQLite FTS5):
  • •python tools/qa_local_search.py --overwrite build
  • •Query (human): python tools/qa_local_search.py search "meta validator" --top 10
  • •Query (agent): python tools/qa_local_search.py --json search "meta validator" --top 10
  1. •Build QA-certified navigation index (Merkle store + posting-list views):
  • •python tools/qa_forensics_cert_index.py build
  • •Views: python tools/qa_forensics_cert_index.py view-keys
  • •Spine example: python tools/qa_forensics_cert_index.py view-get view:cartography/ontology_spine --limit 200
  • •Record example: python tools/qa_forensics_cert_index.py store-get qa_alphageometry_ptolemy/qa_meta_validator.py
  1. •Refresh Results Registry (Layer 4 distillation):
  • •python tools/generate_results_registry.py --control-cutoff-utc-date 2026-01-10

Retrieval workflow (when asked “where is X?”)

  1. •Start with certified spines:
  • •view:cartography/ontology_spine
  • •view:cartography/execution_spine
  • •view:cartography/results_spine
  • •view:cartography/meta_spine
  1. •Use SQLite FTS for precise term search:
  • •python tools/qa_local_search.py search "<query>" --top 20
  1. •When a candidate result is found:
  • •Add/update an entry in Documents/RESULTS_REGISTRY.md with:
    • •claim (1–3 sentences)
    • •evidence path(s)
    • •reproduction command(s)
    • •status (draft/verified/superseded)

Guardrails

  • •Prefer _forensics/ for heavy audit outputs (already ignored by git).
  • •Avoid emitting raw ChatGPT export text; prefer aggregates + file/path references.
  • •If you need to touch “trace-heavy” trees (qa_lab/, vault_audit_cache/, chat_data_extracted/), do it intentionally and record what you did in trace/TRACE_INDEX.md.
  • •Treat pre-2026-01-10 artifacts as legacy until re-validated under trace/ledgered runs.