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
- soulsteads/soulsteads-agent-workboardtheir own version, first seen Nov 15, 2026
- sodejm/copilot-operation-plugin-for-securityword for word, first seen Oct 23, 2026
Other versions2 projects use a version of this.
Word for word · 1
- sodejm/copilot-operation-plugin-for-security★ 2100% same
Edited · 0
None found.
Wrote their own version · 1
- soulsteads/soulsteads-agent-workboard★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository28 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- agent-workboardSkill
- clear-writingSkill
- documentation-impactSkill
- github-state-auditSkill
- gitignore-auditSkill
- milestone-deliverySkill
- repository-checkSkill
- repository-delivery-gateSkill
- repository-orientationSkill
- resumable-worktree-runSkill
- security-reviewSkill
- session-usage-auditSkill
- template-project-creatorSkill
- clear-writingSkill
- documentation-impactSkill
- github-state-auditSkill
- gitignore-auditSkill
- milestone-deliverySkill
- repository-checkSkill
- repository-delivery-gateSkill
- repository-orientationSkill
- resumable-worktree-runSkill
- security-reviewSkill
- session-usage-auditSkill
- template-project-creatorSkill
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
- Distinctive: line 8
Use this skill from a running parent agent to coordinate bounded subagent or resumable work.
- Distinctive: line 27
Run the packaged helper with the active Python interpreter:
- Very specific: line 33
Every command accepts --cwd <project> or --db <exact-path> before the command and --json for machine-readable output.
- Distinctive: line 44
Create a run owned by the current parent session:
- Distinctive: line 55
Put these literal fields in each child prompt:
- Very specific: line 74
Before reporting orchestration complete, run reconcile, inspect status --run-id, read every generated handoff, and resolve all pending, …
- Very specific: line 81
Read the three WORKBOARD_* values from the launch prompt, then atomically claim the assigned task before beginning:
- Distinctive: line 115
If the board is temporarily unavailable, do not fabricate state.
- Distinctive: line 128
Never reassign stale work automatically.
≈ “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.
Files
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.pyEvery command accepts --cwd <project> or --db <exact-path> before the
command and --json for machine-readable output.
Parent workflow
Initialize the board:
workboard.py --cwd <project> --json initCreate a run owned by the current parent session:
workboard.py --cwd <project> --json run start \ --title "<bounded orchestration>" --session-id "<session-id>"- 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.
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 completeortask block.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>- Before reporting orchestration complete, run
reconcile, inspectstatus --run-id, read every generated handoff, and resolve allpending,ready,launched,running,blocked, orstaletasks. 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
pendingtoreadyonly after all prerequisites complete. - Claims use
BEGIN IMMEDIATE; exactly one agent owns a contested task. - Leases last 30 minutes and are renewable. Only
reconcileturns expired running tasks intostale. - Never reassign stale work automatically.
block,complete, andcancelwrite a Markdown handoff automatically.completeandblockaccept repeatable--changed-fileand--validation-filereferences. 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.
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
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.

