Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/shimo4228/claude-harness/session-judgment-miningnpx skills add shimo4228/claude-harness --skill session-judgment-mininggit clone --depth 1 https://github.com/shimo4228/claude-harnessWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/shimo4228/claude-harness/session-judgment-mining)<a href="https://agentmods.dev/skills/shimo4228/claude-harness/session-judgment-mining"><img src="https://agentmods.dev/badge/skills/shimo4228/claude-harness/session-judgment-mining.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00336 | $0.03239 |
| Opus 5 | $0.00168 | $0.01620 |
| Sonnet 5 | $0.00067 | $0.00648 |
| Haiku 4.5 | $0.00034 | $0.00324 |
Grade A, and why
session-judgment-mining scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured today.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 120 lines — stays where its author put it; the contents beside it link to each section on GitHub.
session-judgment-mining — 過去セッション群からの判断・価値観の発掘とスキル化
Purpose: ユーザーの判断・価値観は個々のセッションの修正指示(「ここ直して」「こうじゃない」)として発話され、memory に部分的に残るだけで大半は会話履歴に埋まる。このスキルは過去セッション群を全量発掘し、繰り返された判断を skill として正本化する。learn-eval が「今のセッションから 1 パターン」を抽出するのに対し、これは「過去セッション群の遡及一括発掘」。
初回実施: 2026-07-28、zenn-content repo(33 セッション / 人間発話 314 turn → skill 2 本 + ADR。memory 未記録の判断 8 件を新規発見)。
Step 1: 規模把握と全量/サンプリング判定
対象は ~/.claude/projects/<project-dir>/*.jsonl(<project-dir> は cwd のパスをダッシュ結合したもの。memory/ サブディレクトリは対象外)。
まず人間発話の turn 数で読む量を見積もる。ファイルサイズは判断材料にならない — トランスクリプトの 9 割超は tool_result で、33 セッション 89MB でも人間発話は 114KB だった。
- 人間発話が数百 turn → 全量パス(サンプリング不要。数十 k tokens で通読できる)
- 数千 turn 超 → フィードバック密度の高いセッション(人間 turn 数上位)から読み、キーワード grep(「直して」「違う」「じゃない?」等)で補完
Step 2: 人間発話の抽出(検証済み jq パターン)
主弁別子は origin.kind == "human"。.type=="user" だけで grep すると 9 割が tool_result のノイズになる。.message.content は string と array(画像添付時)の両形がある。
jq -r '
( select(.type=="user")
| select(
((.origin.kind // null)=="human" and ((.isSidechain // false)|not))
or
((.origin // null)==null and (.message.content|type)=="string" and ((.message.content|startswith("<"))|not))
)
| (.message.content | if type=="string" then . else (map(select(.type=="text") | .text)|join(" ")) end)
| select(length>0)
| "=== TURN ===\n" + .
)
' "$f"
- 2 つ目の select 節は compact / resume 後のセッションで
originが付かない人間発話の補完(<local-command-caveat>等の XML ラッパは除外) - 補完 2:
select(.type=="queue-operation") | .contentに enqueue された生プロンプトが入る - セッション開始 timestamp(
head -5 | jq -r '.timestamp')で時系列に並べると判断の変遷(方針の言語化 → 定着 → 例外の発見)が読める。timestamp が取れないファイルがあるので結合後に全ファイルの包含を検算する - 抽出台帳は scratchpad に置く(コミットしない)
Step 3: 通読とテーマ分類
抽出結果を全量通読し、判断・価値観の発話を分類台帳(scratchpad)に落とす。
- 頻出テーマ(複数セッションで反復)と単発を分ける。反復こそ価値観 — 「同じ指摘が 3 セッションで出ている」が正本化の根拠になる
- 発話は引用のまま台帳に残す(要約すると後段で言い回しの証拠力が消える)。各引用にセッション ID を添える
- 拾うのは修正指示だけではない: 方針の言語化(「全体的に方針として〜」)、承認の型(何に GO を出すか)、却下の型(何を却下するか)、メタ習慣(指摘を規約化させる発話)も判断の証拠
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- today Changed daa7f257e187
- 5d ago First seen · 120 lines · 336 tokens per session scan A 2214a7701554
session-judgment-mining is a skill published in the GitHub repository shimo4228/claude-harness (2 stars, last pushed 3d ago), licensed MIT. It adds 336 tokens to every session and 3,239 once invoked, about $0.0017 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
karpathy-llm-wiki
Use when building or maintaining a personal LLM-powered knowledge base. Triggers: ingesting sources into a wiki, querying wiki knowledge, linting wiki quality, 'add to wiki', 'what do I know about', or any mention of 'LLM wiki' or 'Karpathy wiki'.
skills-constitution
当 Agent 接到专业任务(编码/爬虫/文件操作/API调用/数据分析/文档/部署/推送等)时,强制先查记忆层和技能索引,有匹配必用、无匹配必搜、答复时自动推荐(排除已装)。用于防止 Agent 跳过技能直接硬扛通用能力。跨平台通用(WorkBuddy/Claude/ChatGPT/Cursor/Gemini 等 20+ 框架)。完整版本史见 CHANGELOG.md。.
journal
Maintain agent-authored investigation memory over Genomi evidence links, reviewed source findings, decisions, contradictions, and unresolved questions.
context-engineering
Optimizes agent context setup. Use when starting a new session, when agent output quality degrades, when switching between tasks, or when you need to configure rules files and context for a project.
vitaecontext-vitaegraph
Build, deepen, validate, index, and maintain a private hierarchical career knowledge graph from supplied career materials. Use when the user asks to create or update a VitaeGraph, model education with nested courses or thesis work, enrich projects from Git repositories, or supply deep selected career context to…
claude-compaction-restore
Use when a Claude Code session has just compacted, is about to compact, reached context limit, resumed after /compact, or needs to rebuild its working mental model from Claude JSONL transcripts and touched files.