/kb-lint

You are about to run a two-pass lint over the knowledge base at the path provided.

Raw argument: $ARGUMENTS

Validate the input

• Argument is a path to a research run directory (e.g. ./research-runs/<topic-slug>/).
• If empty, ask the user for a path and stop.
• Verify <run-dir>/kb/ exists. If not, tell the user and stop.

Pass 1: mechanical lint (script)

Invoke the plugin's lint tool — schema integrity, edge resolution, orphan args, dangling-evidence args, unreferenced sources, status-stale candidates:

python3 "${CLAUDE_PLUGIN_ROOT}/scripts/kb-tool.py" lint <run-dir>

Output:

• Stdout: structured JSON report (schema findings + graph findings + totals)
• Side effects: writes <run-dir>/kb/.lint-report.json (machine-readable) and appends a lint entry to <run-dir>/kb/log.md

Read the JSON report. Group findings by kind. Mechanical findings are facts about the graph state, not necessarily errors — an orphan arg might be deliberately new; a status-stale-candidate might be correctly weak. Your job in pass 2 is to apply judgment.

Pass 2: semantic lint (you)

Invoke the kb-linting skill. It defines what semantic lint checks for — engagement drift, traceback completeness, entity-page opportunities, contradiction-status drift, central-question coverage. Apply each check across the graph using the mechanical findings as starting points.

For each finding (mechanical or semantic), decide one of:

• Acknowledge — finding is correct, no action needed (e.g. a weak-rated arg is correctly weak; an unreferenced source is intentional background reading).
• Fix immediately — small mutation: add a missing edge, populate a dangling-evidence basis, downgrade a stale status, create a missing entity page, append a note to a source's quality_flags.
• Queue for next research pass — needs more sourcing or research, log as a gap entry in kb/log.md.

Apply fixes via the model's Write/Edit tools. The plugin's hooks will re-index automatically; no manual index updates needed.

Output

After pass 2, append a synthesis to <run-dir>/kb/log.md as a single ## [date] lint | summary entry covering:

• Total findings (mechanical + semantic), grouped by kind
• Number addressed in this lint pass (acknowledged / fixed / queued)
• Open items left for follow-up

Do not write to kb/.lint-report.json directly — it's machine-managed by the script. Do append a narrative entry to kb/log.md with your synthesis.