Guide the interactive parts of a Tron mode run (unattended build of one approved milestone) - prepare the milestone with the user, run the readiness check, show the start command, read the finished report together. Use when the user wants to run a milestone unattended, asks for Tron mode, or a Tron report is ready to read.
2 stars
0 votes
0 copies
0 views
Added November 18, 2026
ai-agentspythongo
Works with
cli
Security analysis
A100/100
Scanned November 18, 2026
$npx -y skills add Akku/stem --skill tron --agent claude-code
Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of Tron?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/akku-tron)
---
name: tron
description: Guide the interactive parts of a Tron mode run (unattended build of one approved milestone) - prepare the milestone with the user, run the readiness check, show the start command, read the finished report together. Use when the user wants to run a milestone unattended, asks for Tron mode, or a Tron report is ready to read.
---
# Tron mode: before and after a run
Rules: `R-TRON` (session duties, role tiers), `R-BATCHING` (approval of a batch), `R-DESIGN` (alignment). Do not repeat them here. The runner is `scripts/tron.py`; this skill never starts a run by itself.
## 1. Prepare the milestone with the user (one question at a time, recommendation each)
1. Milestone and tasks: every task has acceptance criteria and a `size-*` label; list the ready tasks in order.
2. Decision frame (`docs/agent/DECISION_FRAME.md`): show it, adjust for this milestone.
3. Invariants (`docs/agent/INVARIANTS.md`): add project-specific lines.
4. Acceptance scenario: a short file `docs/agent/tron/ACCEPTANCE-<milestone>.md` (or `tron.acceptance_scenario`): what the user would click or run to judge the result.
5. Budget and limits: `max_cost`, `max_hours`, `max_tasks`; evaluator mode (`off|milestone|task`); `pause_after`.
6. Start-and-smoke command: `smoke_cmd` or `preview_cmd`/`preview_url` for the product.
## 2. Commands and models
- Show the configured commands of every role (`worker_cmd`, `evaluator_cmd`, ...; a missing one falls back to `worker_cmd`) and `verify_cmd`. The user confirms them.
- Propose a model per role from the `R-TRON` tier table; model names live in the config. Warn if `CLAUDE_CODE_EFFORT_LEVEL` is set (it beats `--effort`).
- Edit `docs/agent/stem-config.json` only with the user's yes; commit it so the tree is clean.
## 3. Check and start
- Run `python3 scripts/tron.py check --milestone <id>` and fix what it lists (it exits 1 with a list; warnings do not block).
- When it prints `READY`, show the exact `start:` line, ideally `--dry-run` first. The user starts it in a sandbox or with an allowlist; a run costs money, so name the budget again.
## 4. After the run
- Read `docs/agent/tron/REPORT-<run>.md` with the user: Done, Blocked (with the draft questions), evaluator findings, cost, branch `tron/<milestone>`.
- The user decides on merging; the run never pushed, tagged or merged. Answer Blocked questions, then offer to run again (it resumes).
- Defects the user finds go into the tasks as evidence; do not fix them silently.