Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/phj128/autoworker/codenpx skills add phj128/autoworker --skill codegit clone --depth 1 https://github.com/phj128/autoworkerWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00043 | $0.00767 |
| Opus 5 | $0.00022 | $0.00383 |
| Sonnet 5 | $0.00009 | $0.00153 |
| Haiku 4.5 | $0.00004 | $0.00077 |
Grade A, and why
code scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 77 lines — stays where its author put it; the contents beside it link to each section on GitHub.
autoworker:code — Implement One Phase
Called by autoworker:dispatch. Does one thing: implement the first incomplete Phase from the subtask plan.
Execution Flow
1. Read Subtask
Glob `subtask_*.md` (exclude subtask_template.md) →
0 found → stop, prompt to create subtask
1 found → use directly (backward compatible)
multiple → grep `status:` to filter:
- Files without status field treated as active (backward compatible)
- Exactly 1 active → use it
- 0 active → list all files + status, prompt user to choose
- >1 active → report anomaly
→ Read → extract:
- Plan section Phase list (with checkboxes)
- Find the first incomplete Phase (`- [ ]` prefixed Steps)
If subtask doesn't exist or plan is empty → stop, prompt to complete autoworker:subtask-plan first. If all Phases are complete → stop, prompt that autoworker:dispatch should route to autoworker:test.
2. Implement Code
Read all Steps in the Phase, implement one by one:
- Read Step description and target file
- Read target file (if modifying)
- Edit/Write to implement the code change
- After completing the Step, continue to next Step
Executing commands: Steps may involve not just writing code but also executing commands (starting training, running scripts, etc.). Long-running commands use run_in_background=true — execute autonomously, do not ask the user to run manually. After starting background tasks like training, continue to the next step without waiting for completion.
Forbidden:
- Writing code from memory without reading subtask
- Skipping Steps (unless subtask explicitly marks as skippable)
- Checking off boxes yourself (autoworker:checkpoint's responsibility)
- Running tests yourself (autoworker:test's responsibility)
- Making changes beyond the current Phase's scope
- Asking user to manually execute commands that can be done autonomously
3. Output Summary
Phase X code implementation complete:
- Step X.1: <brief change description>
- Step X.2: <brief change description>
→ Invoking autoworker:checkpoint
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 77 lines · 43 tokens per session scan A c6dba0c82732
code is a skill published in the GitHub repository phj128/autoworker (17 stars, last pushed 5mo ago), licensed MIT. It adds 43 tokens to every session and 767 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
flow
Orchestrator for Ticket-Flow — default is --local (all phases in this session with user checkpoints). --parallel works multiple ready tickets at once via worktree-isolated subagents in this session; --serial (one subagent at a time, merge+deploy+cleanup per ticket) and --loop (re-query the ready queue after every…
init
Scaffold a ticket-flow project — one-time setup; trigger on "setz ticket-flow auf", "richte das Projekt für ticket-flow ein", "set up ticket-flow here". Runs bd init with tf's agents template (keeps Claude Code's own memory system), scaffolds spec dirs and worktree settings, migrates an existing KANBAN.md into beads.…
pickup
Internal phase 1 of ticket-flow, normally invoked by ticket-flow:flow — validate Definition of Ready, create (or adopt) the ticket's isolated worktree, set the branch lock, claim the item atomically → In Progress. Invoke directly only for recovery, or when the user explicitly wants just the claim+worktree step. Args…
spec
Draft a spec document for a tracked ticket before implementation — the entry point when the user describes a need in prose: "ich brauche …", "mach die spec", "schreib eine Spec für X", "wir sollten X bauen", "draft a spec". Captures the WHAT plus acceptance criteria and proposes open decisions with recommendations.…
implement
Internal phase 2 of ticket-flow, normally invoked by ticket-flow:flow — execute the plan for the claimed ticket inside its worktree: incremental commits, typecheck/test after each major step. Invoke directly only to continue implementing a ticket that is already picked up.
publish
Everything that makes this repository more public than it is — and the inspection that must run first. Report the publication state (local, private, public), run the offline preflight over the whole history, perform the guarded transition to public, or create the remote repo in the first place (first-time gh repo…