agentsclimarketplace

Agent boundary

Skill danieltty/agent-boundary-protocol/skills/agent-boundary

A lightweight ownership protocol for coordinating Codex, Claude Code, and other coding agents without file-scope conflicts.

Install
npx -y skills add danieltty/agent-boundary-protocol --skill agent-boundary

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when coordinating multiple coding agents, subagents, worktrees, dynamic workflows, or parallel coding sessions and needing explicit file ownership, leader-only integration files, diff validation, worker prompts, or merge gates.

SKILL.md

3.7 KB, 738 tokens by cl100k_base, as published. Nobody here has run it

Agent Boundary

Use this skill before and during parallel coding work. The goal is not to orchestrate agents. The goal is to prevent unclear ownership, same-file edits, and unsafe integration.

Core Workflow

  1. Define the batch goal and success criteria.
  2. Create or update an ownership manifest before workers start.
  3. Assign each worker a disjoint write scope.
  4. Reserve integration hotspots as leader_only.
  5. Run the bundled checker before spawning workers.
  6. Generate worker prompts from the manifest.
  7. Validate each returned diff against the worker scope.
  8. Let the leader integrate leader-only files after worker work returns.
  9. Freeze and re-scope if ownership drift appears.

Bundled Files

  • scripts/agent_boundary.py: dependency-free manifest and diff checker.
  • templates/ownership.json: starter ownership manifest.
  • references/manifest-format.md: manifest fields and rules.
  • references/claude-code.md: Claude Code hook and workflow guidance.

Load a reference only when needed.

Create A Manifest

Prefer a small JSON manifest in the target repo, such as:

.agent-boundary/ownership.json

Start from templates/ownership.json, then edit:

  • batch: short batch name.
  • leader: lead agent/session name.
  • leader_only: files or globs workers must not edit.
  • owners: worker names with owns globs and validation commands.
  • merge_gate: review and test expectations.

If the target repo already has a manifest, update it instead of creating a second competing source of truth.

Check Ownership

From the target repo:

python path/to/agent-boundary/scripts/agent_boundary.py check \
  --manifest .agent-boundary/ownership.json \
  --repo .

Use --repo . when the repo already has tracked files and you want resolved overlap checks against actual paths. Without --repo, the checker still catches invalid fields and exact duplicate patterns.

Generate Worker Prompts

Each worker prompt should include:

  • batch name
  • owned file globs
  • leader-only files
  • deliverable
  • validation commands
  • instruction to stop if a leader-only hook is needed
  • instruction to report unexpected changes in owned files

Keep prompts concrete. Do not ask a worker to "help wherever needed" during a parallel batch.

Validate Returned Work

Before accepting a worker's changes:

python path/to/agent-boundary/scripts/agent_boundary.py diff \
  --manifest .agent-boundary/ownership.json \
  --agent dev-api \
  --from-git main \
  --repo .

Use --files instead of --from-git when validating an explicit file list.

Drift Handling

Ownership drift includes:

  • a worker edited leader-only files
  • a worker edited files owned by another worker
  • two workers claim overlapping file scopes
  • a returned diff contains unowned files

When drift appears:

  1. Stop the affected parallel work.
  2. Preserve the dirty diff if salvage is useful.
  3. Decide whether to integrate, discard, or re-scope the change.
  4. Rewrite the manifest before restarting work.

Do not continue parallel implementation while ownership is ambiguous.

Claude Code

For Claude Code dynamic workflows, agent teams, or subagents, use this skill as the ownership layer. Claude Code can orchestrate agents; this skill supplies the file-scope contract and diff guard. See references/claude-code.md when wiring hooks or writing workflow prompts.

What ships with it: 5 files

11.8 KB alongside SKILL.md, 1 of them executable

agents/

references/

scripts/

templates/

Keep looking

Skills are one crate of 326,984. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.