Skill

fulcrum-start-every-task

Register agent run before touching code. Applies before first Write/Edit/MultiEdit/Bash in session, or when targeted by team invocation.

Install

npx claudepluginhub moabualruz/fulcrum --plugin fulcrum

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Before ANY work in Fulcrum-managed project, call `fulcrum action exec start_agent_run`. Control plane must know agent exists. Non-negotiable — WIP limiter + CoS context builder both read `agent_runs`; run never started cannot be heartbeated, blocked, or completed.

Supporting Assets

agents/openai.yaml

SKILL.md

Similar Skills

cache-components

Guides Next.js Cache Components and Partial Prerendering (PPR) with cacheComponents enabled. Implements 'use cache', cacheLife(), cacheTag(), revalidateTag(), static/dynamic optimization, and cache debugging.

cache-components

139.1k

claude-opus-4-5-migration

2 files

Migrates code, prompts, and API calls from Claude Sonnet 4.0/4.5 or Opus 4.1 to Opus 4.5, updating model strings on Anthropic, AWS, GCP, Azure platforms.

claude-opus-4-5-migration

83.2k

claude-code-plugin-release

1 file

Automates semantic versioning and release workflow for Claude Code plugins: bumps versions in package.json, marketplace.json, plugin.json; verifies builds; creates git tags, GitHub releases, changelogs.

claude-mem

64.7k

Stats

Parent Repo Stars1

Parent Repo Forks0

Last CommitApr 20, 2026

Actions

View Source View Plugin View on GitHub View README

Start every task with start_agent_run

Before ANY work in Fulcrum-managed project, call fulcrum action exec start_agent_run. Control plane must know agent exists. Non-negotiable — WIP limiter + CoS context builder both read agent_runs; run never started cannot be heartbeated, blocked, or completed.

When

User says "implement X", "fix Y", "refactor Z", "investigate Q".
About to call Write/Edit/MultiEdit/Bash first time this session.
Invoked by team (PreToolUse hook set FULCRUM_RUN_ID / FULCRUM_TASK_ID in env — honor those, do not create new).
Transitioned from read-only to mutation mode.

How

# Step 1: workspace_id (no params)
fulcrum action exec get_current_context

# Step 2: start run
fulcrum action exec start_agent_run --json '{
  "workspace_id": "ws_123",
  "task_id": "task_123",
  "agent_role": "software_engineer"
}'

Returns run_id. Keep in scope for every heartbeat_agent_run / complete_agent_run / block_agent_run.

No task_id

fulcrum action exec list_tasks with keyword from user request — check for match.
No match → fulcrum action exec create_task with title + acceptance + owning role. Use returned task_id.
Never guess task_id. Bogus id = policy-layer reject.

Unknown role

fulcrum action exec list_agent_profiles, pick by purpose. Default software_engineer for generic impl; tech_lead for architecture; code_reviewer for review; integration_worker for merges.

Red flags

Called Write/Edit without starting run → stop, start_agent_run, redo edit so tool call logs against correct run.
Multiple run_id in flight same session → block_agent_run stale one before new work.
WIP-limit error → do not loop. See workspace-status-on-session-start to diagnose budget holder.

Worked example

User: "fix the failing auth test in packages/auth/src/tests/login.test.ts".

# 1. Discover workspace context
fulcrum action exec get_current_context
→ { workspace_id: "ws_abc123", project_id: "proj_xyz" }

# 2. Find or create task
fulcrum action exec list_tasks
  workspace_id: "ws_abc123"
  status: "open"
→ { task_id: "task_001", title: "Fix failing auth login test" }

# 3. Start run
fulcrum action exec start_agent_run
  workspace_id: "ws_abc123"
  task_id: "task_001"
  agent_role: "software_engineer"
→ { run_id: "run_999" }

# 4. Proceed with fix — run_id in scope for heartbeat/complete/block