Mohamed Boudra
c5a69a1ad9
Always specify the model in paseo run examples (e.g. --provider codex/gpt-5.4 instead of --provider codex) to prevent agents from launching with wrong defaults.
121 lines
3.9 KiB
Markdown
121 lines
3.9 KiB
Markdown
---
|
|
name: paseo-loop
|
|
description: Run an agent loop until an exit condition is met. Use when the user says "loop", "babysit", "keep trying until", "check every X", "watch", or wants iterative autonomous execution.
|
|
user-invocable: true
|
|
---
|
|
|
|
# Paseo Loop Skill
|
|
|
|
You are setting up a loop — an iterative worker/verifier cycle managed by the Paseo daemon.
|
|
|
|
**User's arguments:** $ARGUMENTS
|
|
|
|
---
|
|
|
|
## Prerequisites
|
|
|
|
Load the **Paseo skill** first. It contains the CLI reference for `paseo loop` and related commands.
|
|
|
|
## Core Model
|
|
|
|
A loop repeats: launch a worker → verify → repeat until done or limits hit.
|
|
|
|
1. **Worker prompt**: what the worker does each iteration
|
|
2. **Verification**: verifier prompt and/or shell checks that judge success
|
|
3. **Sleep**: optional pause between iterations
|
|
4. **Stop conditions**: max iterations and/or max total runtime
|
|
5. **Model selection**: different providers/models for worker vs verifier
|
|
6. **Archive**: optionally preserve agent history after each iteration
|
|
|
|
## Verification
|
|
|
|
Every loop needs at least one form of verification:
|
|
|
|
- `--verify "<prompt>"` — a verifier agent judges the worker's output
|
|
- `--verify-check "<command>"` — a shell command that must exit 0 (repeatable)
|
|
- Both can be combined: shell checks run first, then the verifier prompt
|
|
|
|
## Model Selection
|
|
|
|
Choose the right provider/model for worker and verifier independently:
|
|
|
|
- `--provider <provider/model>` — sets the worker (e.g. `codex/gpt-5.4`)
|
|
- `--verify-provider <provider/model>` — sets the verifier (e.g. `claude/opus`)
|
|
|
|
Default: both use Claude/sonnet. For implementation loops, use Codex for the worker and Claude for the verifier — each catches the other's blind spots.
|
|
|
|
## Archive
|
|
|
|
`--archive` preserves worker and verifier agents after each iteration instead of destroying them. Use this when you need to inspect conversation history for debugging.
|
|
|
|
## Defaults by User Intent
|
|
|
|
### Babysit / watch / check every X
|
|
|
|
```bash
|
|
paseo loop run "Check PR #42. Review CI, comments, and branch status. Fix issues as they arise." \
|
|
--verify-check "gh pr checks 42 --fail-fast" \
|
|
--sleep 2m \
|
|
--max-time 1h \
|
|
--name babysit-pr-42
|
|
```
|
|
|
|
### Keep trying until tests pass
|
|
|
|
```bash
|
|
paseo loop run "Run the test suite, investigate failures, and fix the code." \
|
|
--provider codex/gpt-5.4 \
|
|
--verify "Run the test suite. Return done=true only if all tests pass. Cite the exact command and outcome." \
|
|
--verify-check "npm test" \
|
|
--max-iterations 10 \
|
|
--name fix-tests
|
|
```
|
|
|
|
### Implementation loop with cross-provider review
|
|
|
|
```bash
|
|
paseo loop run "Implement issue #456. Make incremental progress each iteration." \
|
|
--provider codex/gpt-5.4 \
|
|
--verify "Verify issue #456 is complete. Check changed files, run typecheck and tests." \
|
|
--verify-provider claude/sonnet \
|
|
--max-iterations 8 \
|
|
--max-time 2h \
|
|
--archive \
|
|
--name issue-456
|
|
```
|
|
|
|
## Managing Loops
|
|
|
|
```bash
|
|
paseo loop ls # List all loops
|
|
paseo loop inspect <id> # Show details and iteration history
|
|
paseo loop logs <id> # Stream logs
|
|
paseo loop stop <id> # Stop a running loop
|
|
```
|
|
|
|
## Your Job
|
|
|
|
1. Understand the user's intent from the conversation and `$ARGUMENTS`
|
|
2. Decide the worker prompt — self-contained, concrete about what to do
|
|
3. Decide verification — shell checks for objective criteria, verifier prompt for judgment
|
|
4. Choose providers/models for worker and verifier
|
|
5. Choose sleep only when the task is polling or waiting on an external system
|
|
6. Add sensible stop conditions
|
|
7. Run `paseo loop run` with the final arguments
|
|
|
|
## Prompt Writing Rules
|
|
|
|
### Worker prompt
|
|
|
|
The worker prompt must be:
|
|
- self-contained
|
|
- concrete about commands, files, branches, tests, PRs, or systems to inspect
|
|
- explicit about what counts as progress this iteration
|
|
|
|
### Verifier prompt
|
|
|
|
The verifier prompt should:
|
|
- check facts, not offer fixes
|
|
- cite commands, outputs, or file evidence
|
|
- be specific about what "done" means
|