handle-on-call

Invoked as /handle-on-call <pasted Slack alert> or /handle-on-call resume [incident-id].

Silence kills trust on-call—ack fast, triage honest, close the loop.

Full loop: workflow.md.

Step 1 — Mode

Input	Mode
resume / resume <id>	Resume
pasted alert / permalink	Full workflow

Empty input (non-resume) → ask once; stop.

Step 2 — Session

Resume: bash <SKILL_DIR>/scripts/resolve-session.sh [incident-id] — null → stop; show path + timeline; confirm before work.

New: bash <SKILL_DIR>/scripts/start-session.sh '<paste>' → JSON paths + has_thread. Include Slack permalink in paste when you have it.

Step 3 — Slack MCP

Per slack-mcp.md. Find Slack MCP in enabled servers; read tool schemas.

Not present → ask user to enable/connect Slack MCP for this agent; stop. Continue only after MCP is available or user explicitly chooses manual paste.

Step 4 — Ack

Per ack.md. Post via MCP before deep triage. Update tracker ack → done.

Step 5 — Triage

Read alert_path; investigate (logs, code, runbooks). Write report_path per output.md. Update tracker triage → done.

Step 6 — Confirm

Present report. Ask once: fix | skip | need-info. Record in tracker. Do not execute until confirmed.

Step 7 — Execute

Choice	Action
fix	Minimal fix and/or runbook steps; summarize in tracker
skip	Record reason; no code changes
need-info	Jump to Step 9 with need-info reply; stop

Step 8 — Push

Skip if no code changes. Commit + push per push.md. Update tracker push.

Step 9 — Reply

Closing thread message per reply.md via Slack MCP. Update tracker reply → done; meta.md status → closed.

Manual paste fallback only if user chose it in Step 3. End with session path.

Missing jq → offer install.