turqoisehex

configure-context-awareness

Configure the cc-context-awareness context window warning system. Use when the user wants to change context warning thresholds, messages, status bar appearance, or other cc-context-awareness settings.

cleanup

End-of-session housekeeping. Inventories state, commits work, cleans artifacts, documents remaining work. Use when ending a session normally with plenty of context remaining. Lighter than /cold and /finalize.

cold

Cold start preparation for context handoff. Default mode (slim) edits CT in place — parent-direct, no dispatch, ~1% context burn. `/cold full` adds backlog/plan reconciliation, transcript orphan scan, and grill. Use slim mid-session at 80%+; use full at session end.

self-test

Verify cc-sentinel installation integrity. Checks hooks, commands, skills, references, templates, and CLAUDE.md rules. Run after installation or when things seem broken.

audit

Sprint start: spec integrity checking, staleness scan, dependency verification, and initial CT setup. Phase /1 of the sprint pipeline. Also invoked as /1.

build

Automated build execution from approved plan. Routes tasks by classification (Opus/Sonnet/Parent), commits at logical boundaries. Phase /3 of the sprint pipeline. Also invoked as /3.

design

Design and planning phase: brainstorming, task classification (Opus/Sonnet split), decision externalization, and Sonnet prompt generation. Phase /2 of the sprint pipeline. Also invoked as /2.

finalize

Sprint finalization: transcript mining, Sonnet review, accumulated corrections, manual test queue, and channel reset. Phase /5 of the sprint pipeline. Also invoked as /5.

opus

Set channel identity for an Opus session. Detects channel infrastructure, creates CT file, starts Opus listener (receives prompts from orchestrator), starts heartbeat watcher for Sonnet listener. Use as /opus N where N is the channel number.

perfect

Post-implementation quality pass: evaluate, grill loop, verification squad, and proof of correctness with user gates. Phase /4 of the sprint pipeline. Also invoked as /4.

prove

Report-only completion-proof. Runs the shared /5 9-lens adversarial finderSet over the current scope and surfaces survivors read-only (never fixes). Thin alias to /5's finderSet. Also: /prove --selftest for the in-harness materialization smoke-test.

rewrite

Ground-up rewrite of a subsystem informed by hindsight. Not refactoring — a fresh implementation. Inventories, extracts requirements, designs, builds alongside, proves equivalence, swaps.

sonnet

Verification listener service loop. Waits for work files in _pending_sonnet/ directory, executes tasks (commit-verification, squad, implementation), returns to waiting. Infinite loop — never initiates work.

spawn

Launch multiple Claude Code sessions with model selection (opus/sonnet/duo), channel routing, and terminal tab management. Use when the user says /spawn, "launch sessions", "start duo", or wants to open multiple CC windows.

verify

Launch verification squad (5 Sonnet minimum + Codex interleaved when available)

status

Quick session status report showing sprint, phase, progress, blockers, and context usage. Use for orientation checks or when asked about current state.

mistake

Correct and encode a mistake into CLAUDE.md accumulated corrections. Use after any caught mistake — by user, verification, or hook. Searches for existing rules, strengthens or adds new ones.

prune-rules

Review and prune accumulated corrections in CLAUDE.md. Use at soft cap (15 corrections), when rules feel stale, or when upgrading to a new model. User decides each item.

grill

Adversarial self-check after completing any unit of work. Asks four questions (where does this break, what haven't I checked, what's most likely wrong, what assumption is unverified), verifies every checkable answer, and resolves or flags issues. Use after completing a task and before claiming done.