| name | signal-triage |
| description | Convert Slack, Linear, GitHub, or Markdown inbound signals into a small set of agent-ready tasks without opening noisy applications manually. |
Signal Triage Skill
Purpose: protect the human's attention. You are not trying to ingest everything. You are trying to surface only asks that deserve action.
Inputs
Use whichever sources are available:
- MCP tools named
slack, linear, or github.
.agent-harness/signals/latest.json produced by sync_external_signals.py..agent-harness/signals/inbox.md for pasted messages.- Existing
AGENT_TASKS.md.
Classification
Classify each inbound item as:
ignore: FYI, duplicate, social, non-actionable.needs-human: requires judgment, politics, commitment, or sensitive context.needs-spec: likely useful, but missing acceptance criteria or verification.agent-ready: specific enough to dispatch to an isolated worktree.
Output contract
Append only concise task cards to AGENT_TASKS.md:
- [ ] TASK-123 | agent-ready | branch: agent/task-123 | title: Short imperative title
- source: slack|linear|github|manual:<link-or-id>
- goal: Observable result.
- non_goals:
- Explicitly out of scope.
- acceptance:
- User-observable criterion.
- verification:
- Exact command.
- risk:
- Main failure mode.
Hard rules
- Deduplicate against existing tasks before appending.
- Do not dispatch vague tasks.
- Never create more than 5 new tasks in one triage pass.
- Prefer one boring
agent-ready task over five ambiguous tasks.
- If Slack/Linear/GitHub tools require authentication, stop and write the exact missing setup step.
