agent-harness-kit

List specification headers from docs/specs without loading bodies.

Read one specification body by slug with an explicit offset and limit.

List related specification headers and edges without their bodies.

Create a validated specification in docs/specs from structured fields.

Update validated metadata fields without hand-editing frontmatter.

Replace a specification body while preserving its validated frontmatter.

Change a specification status while enforcing source approval rules.

Create a bidirectional validated relationship between two specifications.

Remove a relationship and its inverse from two specifications.

Validate all docs/specs frontmatter, links, and source references.

Start a new action for a task. Returns an actionId.

Record a section in an action. Standard sections: result, tools_used, blockers, next_steps.

Close an action with a one-line summary.

Full task action history, including every action and section. Potentially large; use only for audit or diagnosis.

Compact newest-first action index. Use to discover actions before reading a specific action or section.

Get one action and a compact index of its sections; section contents are not included.

Compact newest-first section index for one action. Filter by section types without loading content.

Read one section content with an explicit character range. This is the only generic action reader that returns long text.

Write a validated, recipient-directed, bounded handoff for a completed action to resume work without loading the full history.

Get the newest completed canonical handoff addressed to a recipient. Returns HANDOFF_NOT_FOUND instead of falling back to history.

List tasks, optionally filtered by status. Excludes archived tasks by default.

Atomically claim a pending task. Returns task_already_claimed if another agent got it first.

Change the status of a task.

Search the project docs folder for content matching a query.

Record one or more files touched during an action, atomically (all-or-nothing). This is the only way to populate the files-touched count shown in the dashboard. Batch every file from a step of work into a single call — a single-element array is correct when only one file was touched.

Mark an acceptance criterion as met. Use the criterion id from tasks.get.

Given a taskId, returns all acceptance criteria for that task with their id, task_id, criterion text, and met status. Use the returned id values to call tasks.acceptance_update(criterionId).

Create a new task in the harness. Use this when the user describes work in natural language. Infer slug, title, description, and acceptance criteria from the conversation. Ask for missing critical info before calling.

Record one or more tool calls made during an action, atomically (all-or-nothing). This is the only way to populate the Tools dashboard. Batch every tool call from a step of work into a single call — a single-element array is correct when only one call was made.

Edit an existing task (title, description, acceptance criteria). Omitted fields keep their current values.

Archive a task. Archived tasks are hidden from default views (CLI and dashboard).

Unarchive a previously archived task, restoring it to default views.

Check that a .claude/agents/*.md definition file exists for every role. Returns { in_sync, agents } where each agent is { ok } or { ok: false, reason: "missing_file" }. Agent file CONTENTS are never inspected — they are meant to be customised freely — so this never reports drift, only absence. Run `…

Snapshot current package.json dependencies to .harness/deps-lock.json

Compare current package.json against .harness/deps-lock.json and report changes

Check lib version, agent files, and harness skills sync status. Returns structured JSON.