Claude-skill-registry eld-record-memory-collector

Collect and verify project information for pce-memory with multi-model validation using Codex CLI. Use when gathering project context, onboarding to a new codebase, or building a knowledge base. Triggers on "collect project info", "build knowledge base", "index this project", or "remember this codebase".

install

source · Clone the upstream repo

git clone https://github.com/majiayu000/claude-skill-registry

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/eld-record-memory-collector" ~/.claude/skills/majiayu000-claude-skill-registry-eld-record-memory-collector && rm -rf "$T"

manifest: skills/data/eld-record-memory-collector/SKILL.md

source content

PCE Memory Collector

Collect project information with multi-model verification to ensure accuracy.

Workflow

Loop 1: Collect   →  observe(ttl=7, unverified)
Loop 2: Verify    →  Codex cross-check → observe+extract(verified)
Loop 3: Validate  →  hash/diff check → feedback

Loop 1: Initial Collection

Scan project structure:

scripts/scan_project.py <project_path>

Read key files and record with observe:

observe(
  source_type="file",
  content="<extracted info>",
  source_id="<file_path>",
  ttl_days=7,
  boundary_class="public|internal",
  tags=["unverified", "<category>"]
)

Categories:

project-info

architecture

dependencies

api

config

Loop 2: Multi-Model Verification

Cross-validate with Codex CLI:

mcp__codex-cli__codex(
  prompt="Verify this claim against the file content:
    Claim: <claim_text>
    File: <file_path>
    Content: <file_content>

    Respond: MATCH | MISMATCH | PARTIAL
    Reason: <brief explanation>"
)

Decision Matrix:

Codex Result	Action
MATCH	`observe(..., extract={mode: "single_claim_v0"})` → permanent
PARTIAL	Refine claim, re-verify
MISMATCH	Discard (let ttl expire)

Loop 3: Hash-Based Validation

Run validation script:

scripts/validate_claims.py <project_path>

The script:

Activates all project claims
Computes current file hashes
Compares with stored provenance
Outputs:
```
VALID | OUTDATED | MISSING
```

Send feedback:

feedback(claim_id, signal="helpful|outdated", score)

Quick Start

1. scan_project.py /path/to/project     # Discover files
2. [Manual] Read files, observe         # Loop 1
3. [Codex] Cross-validate               # Loop 2
4. validate_claims.py /path/to/project  # Loop 3

Scripts

```
scripts/scan_project.py
```
- Scan project and output key files
```
scripts/validate_claims.py
```
- Hash-based claim validation

References

```
references/pce_memory_api.md
```
- pce-memory tool reference

品質優先原則（Superpowers統合）

核心原則

Epistemic Humility: 推測を事実として扱わない。
```
unknown
```
と言う勇気を持つ
Evidence First: 結論ではなく因果と証拠を中心にする
Minimal Change: 最小単位で変更し、即時検証する
Grounded Laws: Lawは検証可能・観測可能でなければならない
Source of Truth: 真実は常に現在のコード。要約はインデックス

「速さより質」の実践

要件の曖昧さによる手戻りを根本から排除
テストなし実装を許さない
観測不能な変更を防ぐ

完了の定義

Evidence Ladder目標レベル達成
Issue Contractの物差し満足
Law/Termが接地している（Grounding Map確認）
Link Mapに孤立がない
ロールバック可能な状態

停止条件

以下が発生したら即座に停止し、追加計測またはスコープ縮小：

予測と現実の継続的乖離（想定外テスト失敗3回以上）
観測不能な変更の増加（物差しで検証できない変更）
ロールバック線の崩壊（戻せない変更の発生）