aiskillstore/marketplace / pending/cagdasyurekli/agy-worker
agy-worker
UnverifiedUse when Codex or Claude Code should delegate repository exploration or implementation to Google Antigravity CLI (agy), then review, verify, repair, and deliver the result.
README.md
View on GitHub# agy-worker package
`agy-worker` is an Agent Skill for delegating repository exploration and
implementation to Google Antigravity CLI (`agy`) while keeping acceptance with
the driver. The package includes its complete portable core runtime, so an installed
skill does not need a repository checkout or a network fetch for ordinary workflows.
Maintenance tools require a repository checkout.
Use it when a repository task benefits from delegated discovery or edits and the driver
can independently inspect the candidate and run the relevant project checks. It is
not a general-purpose provider client, a security sandbox, or evidence that worker
output is correct.
## Requirements
- OpenAI Codex CLI or Claude Code
- Bash, Python 3, and Git
- `agy` on `PATH` with provider access for live dispatch
- A branch-backed disposable Git worktree whose complete provider-readable content
has been reviewed and approved
A provider model slug containing `claude` does not establish driver-host support.
## Start from your driver
After installing the skill, start a new driver session and make the scope and checks
concrete. For example:
> Use the agy-worker skill to add parser error-path tests in this repository. Allow
> edits only under tests, verify with the existing parser test suite, and preserve
> useful partial work if a check fails.
Before a live dispatch, the driver must show the owner the exact task and selected
settings beside the content digest; a later provider notice may use a public-safe
summary. Obtain any missing provider-transmission approval, and ensure secrets,
denied paths, and unrelated private files are absent from the disposable worktree.
Installation alone grants none of those permissions.
Prefer selected-content `--provider-scope` for bounded jobs. Whole-worktree dispatch
remains an explicit content-bound exception, and neither the ordinary facade nor the
advanced raw initial launch has an implicit transmission mode.
## Resolve the bundled runtime
`SKILL_ROOT` is the directory containing `SKILL.md`; Claude Code substitutes
`${CLAUDE_SKILL_DIR}` in skill content. Set it explicitly in each Bash call; do not
read a Claude path variable from the environment. Resolve the runtime
instead of assuming a repository path:
```bash
PIPELINE="$(bash "$SKILL_ROOT/scripts/resolve-pipeline.sh")" || exit $?
"$PIPELINE/doctor.sh" --repo /absolute/path/to/target
```
The resolver accepts a complete plugin, an explicit standalone installation marker,
or this folder with its bundled `runtime/`. It fails closed when required components
are missing. `doctor.sh` is an offline prerequisite check; `ready` does not prove
provider authentication or future job success.
For ordinary work, the driver uses the resolved `workflow.sh` facade to preview the
provider-readable path boundary, run an approved workflow, inspect status, and bind
driver-owned verification before finalization. The lower-level dispatcher, lifecycle,
gate, and receipt commands remain advanced recovery surfaces.
## Package guide
- [Skill router](SKILL.md): when to use the skill, workflow selection, dispatch
notices, hard stops, and delivery rules.
- [Project lifecycle and verification](references/PROJECT_LIFECYCLE_AND_VERIFICATION.md):
preview, run/status/finalize, Verification v2, isolated verification copies, and
bounded repair.
- [Security and compatibility](references/SECURITY_AND_COMPATIBILITY.md): provider
transmission, environment, verifier, host, and distribution boundaries.
- [Troubleshooting](references/TROUBLESHOOTING.md): actionable preflight, provider,
lifecycle, candidate, and verification failures.
These package-owned links work in a standalone copy. Repository release notes,
contributor history, and public-site material intentionally remain outside this
runtime guide.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Unverified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |