agentleFS
sodejmsodejm/portable-agent-repository-kit/.claude/skills/agent-workboard/SKILL.md

agent-workboard

Skill1 starsChanged 3 months agoApache-2.0View on GitHub

Coordinate bounded subagent or resumable work through a durable, private per-project SQLite workboard. Use only from a running parent agent when orchestration needs explicit task state and reconciliation.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 6 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill from a running parent agent to coordinate bounded subagent or resumable work.

    line 8
  2. Distinctive:

    Run the packaged helper with the active Python interpreter:

    line 27
  3. Very specific:

    Every command accepts --cwd <project> or --db <exact-path> before the command and --json for machine-readable output.

    line 33
  4. Distinctive:

    Create a run owned by the current parent session:

    line 44
  5. Distinctive:

    Put these literal fields in each child prompt:

    line 55
  6. Very specific:

    Before reporting orchestration complete, run reconcile, inspect status --run-id, read every generated handoff, and resolve all pending, …

    line 74
  7. Very specific:

    Read the three WORKBOARD_* values from the launch prompt, then atomically claim the assigned task before beginning:

    line 81
  8. Distinctive:

    If the board is temporarily unavailable, do not fabricate state.

    line 115
  9. Distinctive:

    Never reassign stale work automatically.

    line 128

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Agent Workboard

Use this skill from a running parent agent to coordinate bounded subagent or resumable work. It supplies durable, project-local task state; it does not replace the parent agent's judgment, complete subagent prompts, repository policy, or final reconciliation.

Storage and safety

  • In a Git project, the board is stored at <git-common-dir>/codex/workboard.sqlite3. Linked worktrees share this board, and Git does not track it.
  • Outside Git, the board is stored beneath the local agent configuration directory, keyed by a hash of the canonical project path.
  • Markdown handoffs live in the board directory's handoffs/ folder.
  • The CLI creates board directories with owner-only permissions where the platform supports them. Use temporary databases only in automated tests.
  • Store bounded contracts, concise progress, validation summaries, and file references. Never store credentials, tokens, personal secrets, full transcripts, raw tool output, or large code and log dumps.

Run the packaged helper with the active Python interpreter:

python3 .agents/skills/agent-workboard/scripts/workboard.py

Every command accepts --cwd <project> or --db <exact-path> before the command and --json for machine-readable output.

Parent workflow

  1. Initialize the board:

    workboard.py --cwd <project> --json init
  2. Create a run owned by the current parent session:

    workboard.py --cwd <project> --json run start \
      --title "<bounded orchestration>" --session-id "<session-id>"
  3. Route every task with the repository's subagent-routing policy, then add its complete contract: goal, allowed scope, inputs, expected output, acceptance checks, selected model and effort, dependencies, worktree, and branch.
  4. Put these literal fields in each child prompt:

    WORKBOARD_DB=<resolved path>
    WORKBOARD_RUN_ID=<run id>
    WORKBOARD_TASK_ID=<task id>

    Direct the child to claim before it works, heartbeat during long work, and finish with task complete or task block.

  5. Launch the subagent according to that environment's supported mechanism, then immediately bind its returned agent identifier:

    workboard.py --db <path> task bind \
      --task-id <task-id> --agent-id <agent-id> --agent-type <type>
  6. Before reporting orchestration complete, run reconcile, inspect status --run-id, read every generated handoff, and resolve all pending, ready, launched, running, blocked, or stale tasks. Never infer completion solely from a child's final message.

Child workflow

Read the three WORKBOARD_* values from the launch prompt, then atomically claim the assigned task before beginning:

workboard.py --db <path> task claim \
  --task-id <task-id> --agent-id <agent-id>

For work lasting more than a few minutes, renew the 30-minute lease and record only a concise checkpoint:

workboard.py --db <path> task heartbeat \
  --task-id <task-id> --agent-id <agent-id> \
  --progress "<short checkpoint>"

Finish exactly once:

workboard.py --db <path> task complete \
  --task-id <task-id> --agent-id <agent-id> \
  --summary "<result>" --validation "<check and result>" \
  --changed-file "<path>" --validation-file "<path>"

or:

workboard.py --db <path> task block \
  --task-id <task-id> --agent-id <agent-id> \
  --reason "<concrete blocker>" --next-action "<one next action>"

If the board is temporarily unavailable, do not fabricate state. Preserve the run and task IDs in the child prompt or handoff, report the failure to the parent, and retry only safe idempotent operations.

State rules

  • States: pending, ready, launched, running, blocked, completed, cancelled, stale.
  • Dependencies move from pending to ready only after all prerequisites complete.
  • Claims use BEGIN IMMEDIATE; exactly one agent owns a contested task.
  • Leases last 30 minutes and are renewable. Only reconcile turns expired running tasks into stale.
  • Never reassign stale work automatically.
  • block, complete, and cancel write a Markdown handoff automatically.
  • complete and block accept repeatable --changed-file and --validation-file references. Put only a small result in --validation.
  • export --task-id <id> --to <repository-path> is the only supported way to copy a handoff into a repository.
  • gc --older-than <days> is explicit; history is never deleted automatically.

Environment integration

This portable skill does not require a particular agent harness or hook system. A harness that supports the helper's optional hook command may inject board and agent details at subagent start; otherwise the parent places those values directly in the child prompt. In either case, the parent remains responsible for launch policy, reconciliation, and repository state.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.