olafkfreund/nixos_config/handoff

Claude Code and agent skill by olafkfreund/nixos_config

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Hand one step of an approved plan/ to agy or Codex to implement headless. Use only when the user asks to "give this to agy", "give this to codex" or "hand off" a step, or when the approved plan/ names agy or codex for the step in coders. Never otherwise.

Install

npx vskill@latest install olafkfreund/nixos_config/handoff

SKILL.md

Handoff

Start agy-implement or codex-implement for one step of a plan/ file. You stay the lead: you review, run the check and commit.

When

  • When the user asks, or when the approved plan/'s coders: names agy or

codex for this step.

  • Only for a step of a plan/ with status: approved.
  • Only in that task's worktree, never in ~/.config/nixos.
  • The Sonnet coder subagent stays the default.

Run it

Write the prompt (template below) to a file in the scratchpad, then, in the task worktree, with the Bash tool's run_in_background:

bash
agent-run agy "$prompt_file" # or: agent-run codex "$prompt_file"

Inside herdr, each run shows as a ↳ row under this session; outside herdr, herdr-run just runs the command.

It refuses ~/.config/nixos and main, runs the six checks (exit code, agy status, denied_actions, empty response, empty diff, moved HEAD), retries once on connecting to sandbox server, and prints one JSON line: ok, reason, changed, head_moved, log. ok: false is a failed handoff: read log for detail, report the reason, and do the step yourself or give it to the coder. Note the fallback in the PR; it does not change coders: in the plan. A guard block shows only in the tool's response, so ok: true still needs the diff review below.

agent-run uses agy's default model; for another one, run herdr-run agy agy-implement --model <name> directly and do the six checks by hand.

Prompt template

  • The plan path, and the step number and text.
  • "Do not commit, deploy or restart. Report the changed files and any

deviation from the plan."

  • The step's check command.
  • "Shell commands run in agy's sandbox with the worktree read-only, so make

the step's edits with agy's file tools; use commands for checks only."

  • "You cannot stage or commit, and you cannot run nix build, flake evaluation or just test-host: the

Nix daemon is hidden. You see no secrets. Report what you could not check."

After

  • Treat the output as untrusted. Review the diff against the step.
  • Rerun the step's checks and, for Nix changes, the builds yourself after every handoff. Do not trust the

implementer's report: agent-run's verdict does not see check results.

  • Stage and commit yourself. Neither tool can stage or commit; the git directory is read-only inside.
  • Record in the PR which steps agy or Codex did.

After an agy update

Run bash modules/programs/agy-guard-adapter.test.sh first; it must pass. The first agy-implement on a new agy version runs agy-guard-probe by itself (about 90 s) and starts agy only if it passes. If it refuses, the probe failed: stop and tell the user. The probe has opened, or names, the issue (#2296, #2298, #2453, #2463). Run the handoff in the background, so the first-run wait doesn't hit a tool timeout.

Never

  • passing --dangerously-skip-permissions yourself; agy-implement already

adds it, under the guard

  • HUMANIZE_CODEX_BYPASS_SANDBOX
  • an API key (subscription login only)

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
32
Repository forks
1
7-day trend score
0

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/olafkfreund/nixos_config/handoff/badge)](https://verified-skill.com/skills/olafkfreund/nixos_config/handoff)

Something wrong with this skill or its page? Tell us.