Tempo Switch

0. Emit SKILL_INVOKE

Post SKILL_INVOKE to the bus before any stateful action.

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

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

Switch the active operational tempo. This changes both your permission profile (what's auto-approved) and your interaction mode (how autonomous you are).

Task

Argument: $ARGUMENTS

Execution

If no argument is provided, print the tempo selection guide below and stop.
If an argument is provided, normalize it to lowercase and match against the 6 tempos.

Copy the matching profile to settings.local.json:

cp ~/.claude/tempos/<TEMPO>.json ~/.claude/settings.local.json

Print the tempo card for the activated tempo.
Adopt the behavioral mode immediately for the rest of this session.

Tempo Selection Guide

Print this when [tempo] is called with no arguments:

OPERATIONAL TEMPOS
==================

  [tempo] recon      Research only. No code changes. Read, explore, report.
  [tempo] pair       Collaborative. Propose before acting. Tight feedback loop.
  [tempo] glide      Rubber-stamp. Execute planned work, report after each step.
  [tempo] sprint     Full autonomy. Confirm plan once, then hands-off execution.
  [tempo] surge      Incident response. All resources on the critical path.
  [tempo] overnight  Async batch. Queue tasks, create PR, review in morning.

Current: [check settings.local.json to determine -- match against tempo profiles]

Tempo Cards

RECON

TEMPO: RECON
Permission tier: readonly (no git writes, no file edits)
Mode: Research only -- explore, analyze, report findings
DO: Read files, search code, run tests, query APIs, web research
DO NOT: Write files, edit code, create commits, push branches
Interaction: Deliver findings as a report. Let the human decide next action.

Behavioral rules for RECON:

You MUST NOT use Write, Edit, or any file-mutating tools
You MUST NOT run git add, git commit, git push, or any git write operations
You MAY read files, search code, run tests (read-only), query web, use gh for reading
Deliver all findings as structured reports
End with recommendations, not actions

PAIR

TEMPO: PAIR
Permission tier: standard (all Bash, no auto-Edit)
Mode: Collaborative -- propose before acting, tight feedback loop
DO: Propose approaches, show diffs before applying, ask for direction
DO NOT: Make large autonomous changes, skip human input on decisions
Interaction: Alternate moves. Agent proposes, human refines.

Behavioral rules for PAIR:

Always propose your approach BEFORE implementing
Show the specific changes you plan to make and wait for approval
For architecture decisions, present options with trade-offs
Keep changes small and incremental -- one logical step at a time
Ask clarifying questions when requirements are ambiguous

GLIDE

TEMPO: GLIDE
Permission tier: standard + Edit (file edits auto-approved)
Mode: Execute with rubber-stamp -- direction is clear, risk is low
DO: Execute planned work, report after each step, move fast on mechanical tasks
DO NOT: Make architecture decisions, change interfaces, refactor beyond scope
Interaction: Agent plans and executes. Human approves the overall direction.

Behavioral rules for GLIDE:

Execute the work directly -- don't over-explain or over-plan
Report concisely after completing each logical unit
If you hit an unexpected decision point, pause and ask
Good for: tests, docs, wiring, hardening, refactoring, lint fixes

SPRINT

TEMPO: SPRINT
Permission tier: full (all Bash + Edit auto-approved)
Mode: Full autonomous execution -- confirm plan once, then hands-off
DO: Plan comprehensively, then execute all phases without stopping
DO NOT: Stop mid-execution for minor decisions, over-consult
Interaction: Human sets the goal. Agent delivers the result.

Behavioral rules for SPRINT:

Present your full plan ONCE at the start, then execute without interruption
Use parallel agents for independent work streams
Ship the simplest thing that works
Run tests and verify before reporting completion
Post updates to the coordination bus for visibility

SURGE

TEMPO: SURGE
Permission tier: full (all Bash + Edit auto-approved)
Mode: Incident response -- all resources converge on one critical issue
DO: Diagnose fast, fix fast, verify fast. Bias toward action.
DO NOT: Work on anything unrelated, gold-plate the fix, refactor
Interaction: Human triages. Agent fixes. Speed over elegance.

Behavioral rules for SURGE:

Focus exclusively on the critical issue -- nothing else matters
Diagnose root cause first, then apply the minimal fix
Run tests immediately after fixing
If the fix works, ship it. Clean up later.
Post SITREP to bus after resolution

OVERNIGHT

TEMPO: OVERNIGHT
Permission tier: full (all Bash + Edit auto-approved)
Mode: Async batch execution -- no human in the loop
DO: Execute all queued tasks, create PRs with summaries, batch results
DO NOT: Send messages, make irreversible infrastructure changes, deploy
Interaction: Human queues tasks before sleep. Agent delivers PRs in morning.

Behavioral rules for OVERNIGHT:

Execute all tasks in priority order
Create a PR for each logical unit of work with clear summary
If blocked on a task, skip it and note the blocker -- don't wait
Run full test suite before finalizing
Post completion summary to coordination bus

After Switching

After copying the profile and printing the tempo card, confirm:

Tempo switched to <TEMPO>. Permission profile updated.
Behavioral mode active for this session.

Skill Chains

Mandatory

None — tempo is a meta-governance skill (Tier 2) that configures the session. No pre-chain needed. (Reclassified 2026-09-03 from "meta-skill" to "meta-governance: config" per meta-skill-doctrine.md — tempo does not invoke/compose other skills at runtime.)

Advisory

After tempo switch: [bus] (STATUS — announce new tempo for fleet visibility)

Authority

T1 (TRUSTED): May switch tempo without restriction
T2 (Active/High): May switch to recon/pair/glide/sprint; surge/overnight require operator approval
T3 (Medium): May switch to recon/pair; glide/sprint/surge/overnight require operator approval
T4 (Probationary): May switch to recon only; all others BLOCKED
Operator: Override any restriction

tempo

Files1

SKILL.mdskills/tempo/SKILL.md