close

Liminal Agents — Close

The substrate has been ingesting all day. /close reads today's signals and runs the bounded agents over the synthesis. The correction loop is the same as /check: they disagree, you correct one, the record grows.

Flow

1. Run the close script

If the user invoked /close from a daemon notification and $ARGUMENTS contains a surfacing_id, pass it along:

node ${CLAUDE_PLUGIN_ROOT}/skills/close/close.js --surfacing-id=<uuid>

Otherwise:

node ${CLAUDE_PLUGIN_ROOT}/skills/close/close.js

Returns JSON with vault_id, signal_count, signal_summary, up to three threads, and the three agent interpretations.

If vault_id is null and reason is no_signals_today, report that and stop.

2. Present the day

Show the user:

• Today — the signal_summary paragraph, verbatim.
• Threads — each thread's label and one-sentence summary.
• Three reads — Architect / Witness / Contrarian, labeled.

Then:

"Which reading is wrong, and why? Your correction enters the vault."

3. Tag the correction

Map the user's reason to one of the nine canonical correction tags. See the /check skill for the full taxonomy: wrong_frame, wrong_intensity, wrong_theory, right_but_useless, right_but_already_known, too_generic, missed_compensation, assumes_facts_not_in_evidence, off_by_layer.

4. Store the correction

node ${CLAUDE_PLUGIN_ROOT}/skills/check/store-correction.js <vault_id> "<agent_name>" "<tag>" "<reason>"

Confirm and end the close.

Voice rules

• No wellness framing. This is not a journal prompt. No "how was your day", no "what are you grateful for".
• Agents hold their jurisdictions. Do not paraphrase them into softer language.
• If the user says nothing landed today, write a correction with too_generic against whichever agent they single out.