Reference: commands

Commands are slash-command entry points invoked directly in a Claude session. All eight listed here are core (non-pack); they ship with the template.

See dependencies for tool installation requirements.

/configure

The configuration concierge for the harness manifest.

Purpose: Applies a configuration change to harness.config.json through the harness’s own tooling, then validates it against harness.config.schema.json and re-runs the gates. Covers packs (scripts/pack-toggle.sh), the reports/docs site surface and optional plugins (scripts/site-toggle.sh), ontologies (defers binding to /ontology-review), and topic/dimension/output/voice/freshness edits. It never hand-rolls what a script already owns, and never reports a change done while a gate is red.

Usage:

/configure [packs|site|ontologies|topics|verify] [<free-form request>]

With no area it surveys the current configuration and suggests next steps.

What it delegates to: harness-configurator agent, which drives scripts/pack-toggle.sh, scripts/site-toggle.sh, scripts/sync-packs.sh, and the /ontology-review command/skill.

Dependencies: harness.config.json, harness.config.schema.json, scripts/pack-toggle.sh, scripts/site-toggle.sh, jq, ajv.

/falsify

Runs the adversarial falsification gate on one or more findings.

Purpose: Acquires the topic-level run lock (scripts/run-lock.sh), then delegates to the falsification-analyst agent in a bounded slice loop. Each slice polls disk state between passes. Verdicts are ordinal: falsified | weakened | survived | inconclusive. One-round rule: the analyst makes one pass per finding; no retry loops.

Usage:

/falsify --topic <id> [--scope <scope>] [--query-budget <N>] [--claim-budget <N>]

--scope accepts all, dimension:<dim>, or finding:<@id>.

What it delegates to: falsification-analyst agent.

Dependencies: scripts/run-lock.sh, scripts/falsify.sh.

/goal-writer

Authors the session goal JSON and /goal prose for a topic.

Purpose: Writes reports/<topic>/goal.json, validates it against schemas/goal.schema.json, and produces the companion /goal prose section. The --reshape flag evolves an existing goal using append-only versioning (SPEC §11): dimensions and checks may be added but not removed. Computes the content-hash goal version (gv-<sha256[:12]>) via scripts/goal-version.sh and resolves carry/stale/gap membership via scripts/resolve-membership.sh. Does not spawn the orchestrator.

Usage:

/goal-writer --topic <id> [--reshape]

What it delegates to: Nothing — writes directly after schema validation.

Dependencies: scripts/goal-version.sh, scripts/resolve-membership.sh, schemas/goal.schema.json, ajv.

/ontology-review

Reviews, enriches, and validates ontology coverage for a topic.

Purpose: Three sequential phases. Phase 1 (deterministic): runs scripts/ontology-review.sh to refresh reports/<topic>/ontology-map.json and report coverage gaps. Phase 2 (enrich, optional): binds the ontology and retro-classifies existing findings. Phase 3 (author, optional): invokes the ontology-manager skill to create or update ontology YAML. The --strict flag fails on any unresolved type.

Usage:

/ontology-review [--topic <id>] [--enrich] [--strict]

What it delegates to: scripts/ontology-review.sh (Phase 1), ontology-manager skill (Phase 3 when authoring is needed).

Dependencies: scripts/ontology-review.sh, jq, yq, ajv.

/resume

Resumes a paused or interrupted research session.

Purpose: Reads the session continuity file for the topic, validates the current goal against schemas/goal.schema.json, runs scripts/reconcile-session.sh to derive a fresh checkpoint from disk (a finding is DONE iff it schema-validates — which requires a verification block — and its verdict is not falsified; a falsified-but-valid finding is left not-done so its dimension is reworked), then re-spawns the orchestrator agent in full mode with the reconciled state.

Usage:

/resume --topic <id>

What it delegates to: orchestrator agent.

Dependencies: scripts/reconcile-session.sh, schemas/goal.schema.json.

/start

Starts a new research session or extends an existing one.

Purpose: Registers the topic in harness.config.json if not present. Supports three modes: full (complete session from scratch), augment (add one or more dimensions to an existing session), and update (refresh stale findings for the current goal version). Phase 2b incorporates any bound ontology into the session context. Delegates all research work to the orchestrator agent. After the orchestrator returns, invokes the readme skill to reconcile the per-topic navigation README.

Usage:

/start --topic <id> [--goal <goal-json-path>] [--augment [<dim>]] [--update]

What it delegates to: orchestrator agent, then readme skill.

Dependencies: harness.config.json, schemas/goal.schema.json, scripts/reconcile-session.sh.

/status

Reports read-only session state for a topic.

Purpose: Reads research-progress.md for the active topic, counts findings on disk, produces a verdict rollup (survived/weakened/falsified/ inconclusive), and reports per-dimension finding counts. No writes; no agents spawned. Safe to run at any point in a session.

Usage:

/status [--topic <id>]

What it delegates to: Nothing — read-only.

Dependencies: research-progress.md, reports/<topic>/findings/*.json, jq.

/topics

Lists the topic registry from the harness manifest.

Purpose: Reads harness.config.json topics[] and presents each topic’s id, title, namespace, and status. Supports --filter <status> to narrow to topics at a given status, and --ids to emit only topic ids (useful for scripting). Simpler than the topics skill — no index rollup, just the manifest view.

Usage:

/topics [--filter <status>] [--ids]

What it delegates to: Nothing — read-only manifest view.

Note: The topics skill (invoked as a skill, not a command) adds live finding and verdict rollup from research-index.json. See core-skills.

Dependencies: harness.config.json.