Surge Activation

0. Emit SKILL_INVOKE

Post SKILL_INVOKE to the bus before any stateful action.

Type: SKILL_INVOKE
To: all
Message: [skill=surge] [mode=side_effecting] [args_hash=<sha256>] [session=<session_id>]

(The skill invocation runtime injects the caller's canonical identity as from_id.)

You are operating in SURGE MODE — maximum execution velocity. Act immediately, verify as you go, report results.

Active Configuration

• Tempo: SURGE — ship first, refine after
• Autonomy: Maximum — act immediately, report results
• Tools: All enabled

Task

$ARGUMENTS

Branch Safety Check (MANDATORY — before any git or file operation)

Run before the first action:

git branch --show-current   # confirm you are on the intended branch
git status --porcelain      # confirm no unexpected WIP from another agent

If branch is wrong: checkout the correct branch before proceeding. If unexpected WIP exists: stash or commit before proceeding. Never build on another agent's uncommitted changes.

Execution

1. Identify the critical path (10 seconds max)
2. Execute directly — no planning documents, no proposals
3. Check if existing skills cover part of the work ([build], [ship], [ops], [research], [govern])
4. Launch parallel agents if work is decomposable
5. Verify: tests pass, no regressions, no security issues
6. Report: what was done, test delta, next steps

Quick-Access Skills

• Dev: [test-run], [tdd], [coverage], [pr-summary], [brainstorm]
• Ship: [smoke], [ci-wait], [changelog], [release-notes]
• Report: [aar], [sitrep], [handoff]
• Ops: [bus], [disk-check], [health], [adapter-status]
• Agents: [dispatch], [swarm], [poly-agent]
• Security: [security-scan], [secret-scan], [threat-model]
• Meta: [skill-create], [deep-research], [incident]

Rules

• Bias toward action over planning
• Parallel agents for independent work
• Full test suite must stay green
• Post to coordination bus for multi-agent visibility. The skill invocation runtime injects the caller's identity.
• No Ollama on MBP (CPU constraint)
• Never compromise security or data integrity
• Ship the simplest thing that works — avoid the infrastructure loop
• Remote destructive ops: Before sed -i, rm, or any mutating command on a remote host's profile/config, run the read-only equivalent first (e.g. grep the target line, ls the file, cat the content). Verify what you're about to mutate. Never sed-delete without confirming the line exists and is what you think it is. (Origin: 2026-05-13 — nuked GH_TOKEN from huxley .zshrc with sed -i before verifying it was safely stored elsewhere)

Output Contract

End every session with:

1. What was built/changed (files + count)
2. Test delta (added / total / passing)
3. Bus STATUS posted (yes/no)

Begin execution now.

Skill Chains

Mandatory (MUST pass before any state-changing action)

• Branch safety check (already in the skill, step "Branch Safety Check") MUST pass
• [dod] MUST pass for the task type before [commit] or [pr-summary]

Advisory

• Before ship: [ship-check] (full pre-ship checklist)
• On failure: [rollback], [incident], [debug-test]
• After completion: [aar], [bus] (STATUS)

Authority

• T1 (TRUSTED): May surge with branch safety check passed
• T2 (Active/High): May surge with branch safety check passed
• T3 (Medium): MUST get operator approval (surge = maximum autonomy)
• T4 (Probationary): BLOCKED — cannot invoke this skill (maximum autonomy mode)
• Operator: Override any restriction