smlflg/ACT-agentcommandotower

commando-tower-v0

Minimal local operator cockpit for agentic projects.

Commands

python tower.py status
python tower.py focus <project>
python tower.py add-decision "<text>"
python tower.py add-run --project X --harness Y --goal Z
python tower.py pause <project>

GH Token Furnace Harness

This repo can be used as a control surface for issue-based agent work. The harness contract is:

Issue -> Branch -> Artifact -> Tests -> PR -> Review

Start with:

• AGENTS.md for repo-level agent instructions
• AGENT_JOB_TEMPLATE.md for issue bodies
• AGENT_RESULT_TEMPLATE.md for required run output
• .github/ISSUE_TEMPLATE/agent_job.yml for GitHub issue creation
• .github/PULL_REQUEST_TEMPLATE.md for PR review boundaries
• docs/GH_TOKEN_FURNACE_HARNESS.md for the full operating model

Optional GitHub label setup:

scripts/ensure-agent-labels.sh

Local verification:

bash scripts/verify.sh

Recommended bootstrap order:

1. Merge the harness bootstrap PR.
2. Run scripts/ensure-agent-labels.sh.
3. Open a scoped issue from .github/ISSUE_TEMPLATE/agent_job.yml.
4. Work on agent/issue-<id>-<slug>.
5. Store run artifacts in artifacts/issue-<id>/.

Files

• tower.yaml stores state.
• focus.md shows current focus.
• decisions.md stores human decisions.
• runs/ stores run notes.

Non-goals

No database. No web UI. No background processes. No autonomous orchestration.