← ClaudeAtlas

knowledge-ledgerlisted

Show which agent-written lines the user has not yet explained, and run an explain-it-back session that pays that debt down. Use when the user asks what they don't understand in their own code, asks to be quizzed on what the agent wrote, says "knowledge ledger" or "knowledge debt", or at the end of a long coding session when they want to check they can still explain the result.
tsurutanmen/claude-plugins · ★ 0 · AI & Automation · score 67
Install: claude install-skill tsurutanmen/claude-plugins
# knowledge-ledger A hook records every line the agent writes with Write, Edit or MultiEdit into `.knowledge-ledger/ledger.jsonl` in the project. A line leaves the ledger when it disappears from the code, or when the user rewrites it themselves. A line is **explained** once the user has described it in their own words. **Debt** = agent-written lines still in the code that the user has not explained. The script is `scripts/kledger.py` in this skill's directory. Standard library only. Call it with `python3` (or `python` on Windows). ## Commands ``` python3 <skill-dir>/scripts/kledger.py status # table, highest debt first python3 <skill-dir>/scripts/kledger.py status --json python3 <skill-dir>/scripts/kledger.py explain FILE --lines A-B # explanation on stdin python3 <skill-dir>/scripts/kledger.py record --git # agents without hooks: lines added since HEAD ``` ## Explain-it-back session 1. Run `status`. Pick the file with the most debt, and in it a block of 5 to 30 agent-written lines that belong together (one function, one branch). 2. Show the block with line numbers. Ask one question: what does this do, and why is it written this way? Do not hint at the answer. 3. **The user writes the explanation. You never write it for them,** never offer a draft to approve, and never paraphrase their words into something better before recording. The ledger measures what the user can say, not what you can. 4. Read the user's explanation against the cod