- Home
- Skills
- Workflow Automation
- Agent Pigeon — Parallel Agent Delegation
Agent Pigeon — Parallel Agent Delegation
Run independent tasks in parallel, inspect their progress, send supported mid-turn instructions, and continue completed conversations with their context intact.
Free
Works with the AI Agents you already use
Agent Pigeon — Parallel Agent Delegation
Example session with this skill installed
Illustrative workflow request — requires installed runners and configured model routes:
Use Pigeon to add subscription billing to this application. First ask Claude Code with my Opus 5.5 configuration to produce a plan and API contract. Then run Codex with my Astra configuration on implementation, a compatible DeepSeek-configured runner on tests, and an independent agent on security review in parallel. Give each agent the plan and an isolated working directory or explicit file ownership.
Collect their results, send review findings back to the original implementation session, and ask the test agent to verify the integrated fixes. If my Hermes scheduler and messaging channel are already configured, send me a progress digest every 15 minutes and flag blockers or decisions. Otherwise explain the missing setup before promising notifications. Finish with actual changes, test outcomes, and remaining issues; do not deploy.
- Read your context and instructions
- Compiled the agent pigeon
Illustrative workflow — not a recorded run
This example describes the expected coordination and report format. No agents have been executed for this listing; no session IDs, test results, or completion claims are fabricated.
Task handoffs
| Stage | Agent assignment | Deliverable / dependency |
|---|---|---|
| Plan | Claude Code with a configured Opus 5.5 route | Subscription states, API contract, acceptance criteria |
| Implement, in parallel | Codex with a configured Astra route | Billing endpoints and webhook handling based on the plan |
| Test, in parallel | Compatible runner with a configured DeepSeek route | Tests for duplicate events, cancellation, and failed payments |
| Review, in parallel | Independent configured agent | Findings on signature verification, permissions, and edge cases |
| Fix | Resume the original implementation session | Address collected findings without starting over |
| Verify | Continue the test session against integrated changes | Actual test outcomes and unresolved failures |
The coordinator passes artifacts between agents and reconciles conflicting findings. Independent work runs concurrently; integration tests wait until there is an implementation to exercise. Concurrent edits use separate directories or agreed file ownership.
Example progress digest
The following is sample wording, not live status:
“Planning is complete. Implementation and test preparation are running in parallel. Review has raised a refund-policy question. Decision needed: revoke access immediately after a refund, or retain access until the billing period ends? This policy remains pending your answer.”
Optional Hermes delivery
A separately configured scheduler can trigger Hermes to query Pigeon and send this digest every 15 minutes through your configured messaging channel. This is an integration pattern, not a built-in Pigeon scheduling feature. Without that setup, request progress on demand; host-supported background waiters can provide completion notifications.
Final handoff format
• Changes implemented, with affected files and relevant session references from the real run.
• Tests actually executed, their outcomes, and any skipped checks.
• Review findings addressed and issues still open.
• Decisions needed from you and proposed next steps.
Model names above are illustrative configurations. Availability and controls depend on the runner, provider, and account; model usage is billed separately.
Connects securely to your tools. The creator never sees your data.
About this skill
One request. Multiple agents. A connected workflow.
Agent Pigeon lets your current agent delegate work to independent Claude Code, Codex, and Hermes sessions through one local CLI. Run independent tasks in parallel, inspect their progress, send supported mid-turn instructions, and continue completed conversations with their context intact.
Example: ship a subscription billing feature
Start with a shared plan, split the work, then reconnect the results. The following is an illustrative workflow, not a recorded execution or a promise of bundled model access.
- PLAN — Have Claude Code, configured with an available Opus 5.5 model, inspect the application and propose the subscription lifecycle, API contracts, and acceptance criteria. The coordinating agent passes the approved plan to each worker.
- RUN IN PARALLEL — Have Codex, configured with an available Astra model, implement the billing endpoints and webhook handling. In a separate working directory, have a compatible runner configured for DeepSeek build tests against the agreed contract. A third independent agent reviews webhook verification, duplicate events, and access-control boundaries.
- CONNECT THE RESULTS — The coordinating agent collects findings and sends actionable feedback to the implementation agent. Resume the same session for fixes, then ask the test agent to validate the integrated changes. Model routing and live steering depend on the selected runner and endpoint.
- KEEP YOU INFORMED — With a separately configured scheduler and messaging channel, Hermes can periodically query Pigeon task state and send a concise progress digest. A 15-minute digest can show completed work, active tasks, blockers, and decisions needed from you.
- HAND OVER — Collect the implementation summary, actual test results, unresolved issues, and next steps. Deployment remains a separate user-authorized action.
Parallel work plus sequential handoffs
Plan → implementation + tests + independent review in parallel → collect findings → resume implementation for fixes → re-test → final summary.
Pigeon provides session and execution management. Your coordinating agent supplies each task with its context and manages handoffs; this skill does not install an autonomous workflow scheduler. Use separate working directories or explicit file ownership for concurrent edits.
Progress updates through Hermes
Illustrative message format
“Billing update: planning is complete. Implementation and test preparation are running in parallel. Review has raised a decision: should a refund revoke access immediately or at the end of the billing period? Waiting for your decision before implementing that policy.”
Scheduled delivery requires an external scheduler, a configured Hermes messaging integration, and access to Pigeon state. The current Pigeon Hermes plugin supports delegation, steering, and on-demand result queries; it does not include a built-in periodic notification service. Notifications in other hosts depend on their background-task completion support.
More workflows
- Incident investigation: agents inspect application code, logs, and recent changes in parallel; the coordinator compares hypotheses, assigns a targeted fix, and requests verification.
- Research to report: agents investigate separate questions, another checks the evidence, and the coordinator assembles the report. A configured Hermes notification workflow can announce progress or completion.
Requirements and what's included
This package includes the original portable SKILL.md, installation documentation, and MIT license. Install Pigeon CLI separately from https://github.com/luokinging/agent-pigeon. Requires Node.js 22.13+, npm, Git, local shell access, and at least one installed and authenticated agent tool. Primarily supports macOS and Linux.
Claude Code, Codex, and Hermes are execution tools; Opus, Astra, and DeepSeek refer to illustrative model choices. Use actual model IDs supported by your configured runner and provider. These routes have not been demonstrated in this listing. Model access, API charges, and subscriptions are not included.
Pigeon is MIT-licensed open-source software. Delegated agents may execute commands and read or modify project files within the task scope you authorize.
Requires pre-installed Pigeon CLI. Background notification depends on host-agent background shell support. Does not manage auth credentials or model access.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 16 days ago
- Free with an Agensi account
Requires
Permissions
Executes the local Pigeon CLI and configured agent runners. Delegated tasks can read/write project files and run commands according to the user-authorized scope. Stores local session/configuration data under PIGEON_HOME (default ~/.pigeon). Runners communicate with their configured model providers using separately configured authentication. Model usage charges are separate.