|
| 1 | +# AGENTS.md |
| 2 | + |
| 3 | +This project uses Oh My OpenAgent with OpenCode. Read this file before planning |
| 4 | +or changing files. |
| 5 | + |
| 6 | +## Orchestrator Contract |
| 7 | + |
| 8 | +- Start by identifying the real user goal and the smallest observable result |
| 9 | + that satisfies it. |
| 10 | +- Delegate independent search, implementation, and verification work to |
| 11 | + background agents when the task has separable parts. |
| 12 | +- Keep direct edits focused on work that does not need specialist context. |
| 13 | +- Verify worker results yourself before using them as evidence. |
| 14 | + |
| 15 | +## Agent Selection Rules |
| 16 | + |
| 17 | +- Use `explore` for fast codebase search, policy lookup, and path discovery. |
| 18 | +- Use `librarian` for docs, external references, and examples from other repos. |
| 19 | +- Use `oracle` for architecture review, risk checks, and debugging analysis. |
| 20 | +- Use `sisyphus` or `sisyphus-junior` for execution work that needs file edits. |
| 21 | +- Use `prometheus` for plans that need user approval before implementation. |
| 22 | + |
| 23 | +## Calling Conventions |
| 24 | + |
| 25 | +- Give each delegated agent one goal, a clear deliverable, and exact paths or |
| 26 | + commands to inspect. |
| 27 | +- Ask read-only agents for findings with file paths and evidence, not edits. |
| 28 | +- Ask implementation agents to include the verification commands they ran. |
| 29 | +- Retrieve background output before marking a delegated task complete. |
| 30 | + |
| 31 | +## Instruction Phrasing |
| 32 | + |
| 33 | +- Prefer affirmative constraints: "Keep scratch files under ./.omo/session-work/" |
| 34 | + instead of broad negative phrasing. |
| 35 | +- State required artifacts directly: evidence files, screenshots, transcripts, |
| 36 | + command output, or PR links. |
| 37 | +- Name the allowed work area when the task must avoid unrelated files. |
| 38 | + |
| 39 | +## Workflow Loop |
| 40 | + |
| 41 | +- Plan the work, then split independent tasks before starting edits. |
| 42 | +- Run the cheapest useful failing check or missing-docs search before the fix. |
| 43 | +- Apply the smallest change that satisfies the request. |
| 44 | +- Run targeted checks and one real surface QA step. |
| 45 | +- Summarize what changed, what passed, and any remaining risk. |
| 46 | + |
| 47 | +## Project Scratch Space |
| 48 | + |
| 49 | +Use `./.omo/session-work/` for temporary clones, downloaded references, logs, and |
| 50 | +agent scratch files. Keep generated evidence under `./.omo/evidence/` when the |
| 51 | +project wants durable QA records. |
0 commit comments