docs/goal-quickstart.md
/goal Quickstart
Run a bounded long-horizon goal in under five minutes.
Create a goal
Pass the objective as a positional argument, with hard budgets and a workspace root. Write the contract to a file with --output so you can review and run it.
picasso goal create "summarize the auth module changes since main" \
--workspace-root . \
--max-wall-time 30m \
--max-cost-usd 2 \
--max-steps 20 \
--max-subagent-spawns 4 \
--output goal.yamlExpected result:
- Picasso writes a goal contract.
- The contract includes objective, scope, budgets, completion criteria, and allowed tools.
- The goal can be validated before running.
Validate the contract
picasso goal validate goal.yamlExpected result:
- Schema validation passes.
- All budget fields are present.
- Completion criteria are concrete.
Run the goal
picasso goal run goal.yamlExpected result:
- The loop enters PLAN, ACT, OBSERVE, and ASSESS phases.
- The TUI
/goalmonitor shows budget, state, current tool, memory writes, and iteration timeline. - If a cap is hit, the goal pauses with a resume token instead of failing silently.
Monitor from the TUI
Inside Picasso:
/goal monitorThe monitor shows:
- Current state: running, retrying, paused, failed, or completed.
- Budget tiles for wall time, cost, steps, and subagents.
- Current tool work.
- Memory writes.
- Resume controls.
Resume
picasso goal resume <goal-id>Use resume after:
- Wall-time cap reached.
- Cost cap reached.
- Provider credit exhausted.
- Crash recovery from a checkpoint.
Good first goal shape
objective: Summarize auth module behavior changes since main.
completionCriteria:
- type: all-tests-pass
testCommand: pnpm test auth
- type: agent-self-assesses-complete
rubric: The summary names behavior changes, risk, and verification.
budget:
maxWallTime: 30m
maxCostUSD: 2
maxSteps: 20
maxSubagentSpawns: 4
scope:
allowedTools: [read_file, grep, glob, exec]
allowedSubagents: [codebase_investigator]
permissionMode: safe
workspaceRoot: .