Commands

All 15 slash commands registered in commands/. Each command is invoked as /command-name in a Claude Code session.

Harness Lifecycle

Commands for initialising, monitoring, constraining, and maintaining the living harness.

/harness-init

  • Skills read: harness-engineering, context-engineering
  • Agents dispatched: harness-discoverer

Set up a living harness for the project. Interactive feature selection walks through context, constraints, garbage collection, CI templates, and observability. Dispatches the harness-discoverer agent to scan the repo and identify the stack before generating HARNESS.md. Safe to re-run — subsequent runs add features incrementally without overwriting existing configuration.

/harness-status

  • Skills read: none
  • Agents dispatched: none

Quick harness health read with no agent overhead. Reads HARNESS.md and cross-references project state to produce a summary of enforcement ratio, drift since last audit, and garbage collection state. Use this for a fast pulse check between deeper audits.

/harness-audit

  • Skills read: none
  • Agents dispatched: harness-discoverer, harness-auditor

Full meta-verification of the harness. Dispatches the harness-discoverer agent to scan the repo, then the harness-auditor agent to compare what HARNESS.md claims against what the project actually contains. Reports mismatches, stale constraints, and missing enforcement. Use this after significant changes to the project structure or CI pipeline.

/harness-constrain

  • Skills read: constraint-design, verification-slots
  • Agents dispatched: none

Add a new constraint or promote an existing one. Interactive — asks what rule you want to enforce, helps design the constraint (including choosing between agent-scoped and deterministic enforcement), and writes it into the Constraints section of HARNESS.md. Also configures the verification slot if deterministic enforcement is selected.

/harness-gc

  • Skills read: garbage-collection
  • Agents dispatched: harness-gc

Manage and run garbage collection rules. Two modes:

  • Add: Define a new periodic GC rule — what to check, how often, and whether it needs agent judgement or can be deterministic.
  • Run: Execute existing GC rules on demand, outside their normal schedule.

Dispatches the harness-gc agent to perform rule evaluation.

/harness-health

  • Skills read: none
  • Agents dispatched: harness-auditor (Deep mode only)

Generate a comprehensive health snapshot. Two modes:

  • Quick: Reads existing data only — no agents dispatched. Produces enforcement ratio, mutation trends, learning velocity, cadence compliance, and meta-observability status.
  • Deep: Dispatches the harness-auditor agent for a full audit before generating the snapshot. Use this for scheduled health checks.

Snapshots are saved to observability/snapshots/ with a datestamped filename.

Assessment & Improvement

Commands for evaluating AI literacy and aggregating assessments across repositories.

/assess

  • Skills read: ai-literacy-assessment
  • Agents dispatched: none

Run a full AI literacy assessment against the ALCI framework. Scans the repo for evidence of literacy practices, asks clarifying questions where evidence is ambiguous, and produces a timestamped assessment document. After assessment, applies immediate habitat fixes, recommends workflow changes, captures a reflection, and adds a literacy level badge to the project README.

/portfolio-assess

  • Skills read: portfolio-assessment
  • Agents dispatched: none
  • Flags:
    • --local <path> — scan repos under a local directory
    • --org <github-org> — discover repos from a GitHub organisation
    • --topic <tag> — filter repos by GitHub topic

Multi-repo assessment aggregation. Discovers repositories using the specified source, gathers individual assessments, and produces a portfolio view with level distribution, shared gaps, outliers, and a prioritised improvement plan grouped by organisational impact.

Habitat Setup

Commands for bootstrapping and monitoring the complete AI Literacy habitat.

/superpowers-init

  • Skills read: harness-engineering, context-engineering
  • Agents dispatched: harness-discoverer

Bootstrap the full AI Literacy habitat in eight steps:

  1. Discover the stack
  2. Generate CLAUDE.md
  3. Generate HARNESS.md
  4. Generate AGENTS.md
  5. Generate MODEL_ROUTING.md
  6. Generate REFLECTION_LOG.md
  7. Scaffold CI templates
  8. Produce initial health snapshot

Safe to re-run — existing files are preserved and only missing components are added.

/superpowers-status

  • Skills read: none
  • Agents dispatched: none

Full habitat health dashboard. Checks every component of the AI Literacy habitat and reports status per section:

  • Habitat files — presence of CLAUDE.md, HARNESS.md, AGENTS.md, MODEL_ROUTING.md, REFLECTION_LOG.md
  • Harness enforcement — constraint count and enforcement ratio
  • Agent team — agent definitions and availability
  • Compound learning — reflection entries and curation state
  • Model routing — routing table and cost data
  • CI status — workflow presence and recent run health

Each section reports OK, WARNING, or MISSING.

Workflow

Commands for day-to-day development workflow support.

/reflect

  • Skills read: none
  • Agents dispatched: none

Capture a post-task reflection. Appends a structured entry to REFLECTION_LOG.md. Asks three questions:

  1. What was worked on?
  2. What was surprising?
  3. What should future agents know?

Classifies the signal type (technique, constraint, tooling, process) so that reflections can be filtered and curated later.

/cost-capture

  • Skills read: cost-tracking
  • Agents dispatched: none

Capture AI tool cost data for the current period. Finds the previous cost snapshot, guides you through provider dashboards to collect current spend and token usage, records the data, compares against the previous period, and updates MODEL_ROUTING.md with observed cost trends.

/extract-conventions

  • Skills read: convention-extraction
  • Agents dispatched: none

Guided convention extraction session. Surfaces tacit team knowledge through five structured questions covering naming, error handling, testing, architecture, and code style preferences. Maps answers to concrete CLAUDE.md conventions and HARNESS.md constraints. Use this when onboarding AI to an existing codebase or after team composition changes.

/convention-sync

  • Skills read: convention-sync
  • Agents dispatched: none

Sync HARNESS.md conventions to other AI coding tools. Reads the Context and Constraints sections of HARNESS.md and generates tool-specific convention files for Cursor, Copilot, and Windsurf. Ensures all AI coding tools in the team share the same project rules regardless of which editor is used.

/worktree

  • Skills read: none
  • Agents dispatched: none

Manage git worktrees for parallel agent isolation. Three modes:

  • /worktree spin [name] — Create a new isolated worktree branched from the current HEAD. Use this to give a sub-agent its own working directory without interference.
  • /worktree merge [name] — Merge the named worktree back into the current branch.
  • /worktree clean [name] — Remove the named worktree and its branch.