cairn-attention

Skill: cairn-attention

You are surfacing Cairn's pending-attention queue inline so the operator can resolve drafts and findings without leaving the chat. Spec: docs/PLUGIN_ARCHITECTURE.md §11.

Trigger gate

Only fire when:

• The most recent SessionStart additionalContext flagged attention_count > 0, OR
• The operator typed /cairn-attention (escape hatch — slash command not yet wired; for now treat any "show me pending" / "what's in the inbox" message as a manual trigger).

Skip when:

• A cairn-direction task is in flight for this session.
• The operator's prior turn was [c] "later" on this skill — wait until the next session.

Step 1 — read attention sources

Run these in parallel. Use the MCP tools exclusively for DEC content; never cat, Read, or otherwise inline-read draft files — that wastes thousands of tokens on body text the operator never sees.

1. List draft paths only (no contents): Bash: ls .cairn/ground/decisions/_inbox/*.draft.md 2>/dev/null
2. For each path, call cairn_decision_get({path: "<path>"}) — returns parsed {id, title, source_file, status, capture_source}. Use that to build the surface tuple. Do not Read/cat the draft body.
3. Latest baseline audit (path only): Bash: ls -1t .cairn/baseline/sensor-audit-*.yaml | head -1
4. Drift events: cairn_search({query: "drift"}) against the staleness log if any.
5. Recent invalidation events: read the per-session events marker, then list .cairn/events/*.json newer than last_polled_ts.

For each item, build a tuple {kind, id, title, source, severity} from the MCP responses.

Step 2 — sort and cap

Sort by:

1. Hard inconsistencies first (kind=conflict)
2. DEC drafts next (kind=decision_draft) — newest first
3. Baseline findings (kind=sensor_finding) — by sensor severity
4. Drift events (kind=drift)
5. Cross-session invalidation events (kind=invalidation)

Surface at most 3 items per turn. After three picks, prompt:

3 resolved. Continue with the next batch? [a] yes [b] later

Step 3 — surface each item as A/B/C

For each item, render an inline question. Examples:

DEC draft:

DEC-0042 (draft) — "Switch payment processor from Stripe to Adyen" source: docs/billing.md ingestion · proposed: 2026-05-04 [a] accept (move to canonical) [b] reject (delete draft) [c] edit before accepting

Baseline finding:

Sensor stub_catalog_hits flagged 4 violations in services/auth/ — stub strings ("TODO", "FIXME") in production paths. [a] triage now (open file) [b] accept as baseline (suppress) [c] defer

Invalidation event:

A modified DEC-0019 (which you're using). Source: another session at 2026-05-04T20:14. [a] refresh in-scope [b] continue under old [c] abort current task

Use AskUserQuestion with the labels. After the operator picks, call the resolver:

cairn_resolve_attention({kind: "decision_draft", item_id: "DEC-0042", choice: "a"})

The tool dispatches by kind: decision_draft for accept/reject/edit, baseline_finding for triage/suppress/defer, invalidation_event for refresh/continue/abort, bypass and review for Stop-hook surfaces. On decision_draft + a, the tool also strips the originating source comment and replaces it with a // See DEC-NNNN citation when the DEC came from init-source-comments.

Step 4 — stamp the events poll cursor

After draining cross-session invalidation events, advance the per- session marker so the next Stop hook poll only sees newer events:

Bash: node -e "const x = require('@isaacriehm/cairn-core'); x.stampEventsPoll({repoRoot: process.cwd(), sessionId: process.env.CLAUDE_SESSION_ID, ts: Date.now()})"

(The stampEventsPoll runtime helper lives in cairn-core/session; the Stop hook also calls it on every assistant turn end. Calling it here keeps the cursor fresh after the operator drains attention.)

Hard rules

• Surface ≤ 3 items per turn. Do not flood the chat.
• Every option must cite the underlying source (file path, sensor id, session id) so the operator has full context.
• Never auto-resolve. Even soft conflicts route through A/B/C.
• Hard inconsistencies (kind=conflict) block the next cairn-direction invocation until resolved — make that visible in the surface text.
• Caveman-ultra style for chat replies; full English in any DEC body the skill writes.