| name | slate-run |
| description | Fast token-efficient execution loop for active repo work. Useful for Project X / AG Trade style workflows where agents must stay scoped, safe, and brief.
|
slate-run
Fast execution loop for real projects.
Mission
Maximize useful progress per token during active repo work.
Use for Project X, AG Trade, repo maintenance, bugfix loops, dashboard edits, risk-control changes, and any task where the agent should avoid broad scanning or long narration.
Loop
- Restate goal in one line.
- Inspect only relevant files first.
- Make one focused change.
- Run smallest relevant check first.
- Run broader check only when needed.
- Report compact proof.
- Suggest next prompt/command.
Output
Goal:
Files:
Action:
Check:
Result:
Next:
Efficiency rules
- No full repo scan unless task requires it.
- No architecture essay before patch.
- No repeated context user already provided.
- No long logs; quote shortest decisive line.
- No multi-feature patch unless user asks.
- Prefer targeted tests before full suites.
- If blocked, state blocker + exact next action.
Safety boundary
For trading, payments, auth, secrets, deploys, destructive DB ops, or real capital:
- Stop before live execution.
- State boundary clearly.
- Offer dry-run, simulation, manual approval gate, or test harness.
- Do not add wallet keys, private keys, real order placement, or automatic capital deployment.
Example
Goal: add rejected-trade journal tile.
Files: dashboard.py, btc5m_journal.py.
Action: added rejected_count + reason summary.
Check: `python cli.py verify` passed.
Result: dashboard shows fills + rejects.
Next: add reject-rate trend chart.