# delegate-skills
[](https://github.com/amElnagdy/delegate-skills/actions/workflows/relays.yml)
[](https://www.skills.sh/amElnagdy/delegate-skills)
[](LICENSE)
**Create your fleet of lanes. One orchestrator, the right implementer for every job.**
Discover the implementer CLIs already installed on your machine, organize them into lanes like
`feature`, `tests`, and `ui`, then delegate by lane — or choose one implementer directly. Either way,
you keep the review and the commit.
```bash
npx skills add amElnagdy/delegate-skills
```
Then ask your orchestrating agent to create the fleet:
```text
Use $delegate-setup to discover my installed implementer CLIs and create a fleet for feature, tests, and UI work.
```
Or delegate directly:
```text
Use $codex-delegate to have Codex implement the refactor in services/billing/, then review and commit it.
```
```mermaid
flowchart LR
S["$delegate-setup<br/>discover → propose → approve"] --> F["Example fleet"]
O["Your orchestrator"] --> F
F -->|"feature"| A["OpenCode"]
F -->|"tests"| B["Codex"]
F -->|"ui"| C["Cursor"]
A --> R["Review the diff<br/>Run the gates"]
B --> R
C --> R
R --> L["You land the commit"]
```
## Choose how you delegate
### Create a fleet
| Skill | Job |
| --- | --- |
| [`delegate-setup`](skills/delegate-setup/SKILL.md) | Discover installed CLIs, propose **fleet lanes**, and write global or project config after you approve. Never dispatches work. |
A **fleet** is your set of named lanes. Each **lane** binds a kind of work to one implementer and
optional dials such as model, effort, or variant. Setup discovers what is available, proposes a compact
fleet, shows you the complete configuration, and writes only after explicit approval.
Configuration can apply globally or to one repository. Once it is ready, dispatch with the matching
`*-delegate` skill and `--lane <name>`. Explicit flags override lane dials, and the wrong implementer
skill for a lane fails loud. Project config is content-bound to explicit setup approval, so cloned or
edited project lanes fail closed until re-approved. See the
[`delegate-fleet.v1` schema](skills/delegate-setup/references/schema.md) for paths, supported dials,
and overlay behavior.
### Delegate directly
Skip setup when you want one implementer or one-off dials. Pick the skill for a CLI you have:
| Skill | Implementer CLI | Write access (default) | Read-only run | Resume |
| --- | --- | --- | --- | --- |
| [`aider-delegate`](skills/aider-delegate/SKILL.md) | [Aider](https://aider.chat) (`aider`) — any OpenAI-compatible endpoint, including a local or self-hosted model via `--api-base` | `--yes-always` with `--no-suggest-shell-commands`; no sandbox or permission modes; commits force-disabled [^aider] | `--read-only` (`--dry-run`) | `--resume-last` (chat history, per-worktree) |
| [`agy-delegate`](skills/agy-delegate/SKILL.md) | Google Antigravity (`agy`) | Antigravity's own `permissions`; bypass opt-in | `--read-only` (`plan` mode) | `--resume-last`, `--conversation <id>` |
| [`claude-delegate`](skills/claude-delegate/SKILL.md) | [Claude Code](https://code.claude.com/docs/en/overview) (`claude`) | `acceptEdits` + explicit tool surface | `--read-only` (`plan` mode) | `--resume-last`, `--session <id>` |
| [`cline-delegate`](skills/cline-delegate/SKILL.md) | [Cline](https://github.com/cline/cline) (`cline`) | `--auto-approve true` in act mode; upstream sandbox not configured by the relay | `--plan` + `--auto-approve false` (relay-enforced pair) | — (headless JSON resume unsupported) |
| [`codex-delegate`](skills/codex-delegate/SKILL.md) | [OpenAI Codex](https://github.com/openai/codex) (`codex`) | `--sandbox workspace-write` | `--read-only` | `--resume-last`, `--session <id>` |
| [`commandcode-delegate`](skills/commandcode-delegate/SKILL.md) | [Command Code](https://commandcode.ai/docs/headless) (`cmd`; `cmdc` on Windows) | `--yolo` — the only headless write state; no sandbox [^commandcode] | `--read-only` (withheld tools + `plan`) | `--continue-last`, `--session <id>` |
| [`cursor-delegate`](skills/cursor-delegate/SKILL.md) | [Cursor Agent](https://cursor.com/cli) (`cursor-agent`) | `--force`; `--no-force` withholds command approval | `--read-only` (plan mode) | `--resume-last`, `--session <id>` |
| [`grok-delegate`](skills/grok-delegate/SKILL.md) | Grok Build (`grok`) | workspace-scoped; `--full-access` opt-in | `--read-only` — best-effort [^grok] | `--resume-last`, `--session <id>` |
| [`kimi-delegate`](skills/kimi-delegate/SKILL.md) | [Kimi Code](https://moonshotai.github.io/kimi-code/en/) (`kimi`) | `auto permission mode`, always | — [^none] | `--resume-last`, `--session <id>` |
| [`muse-delegate`](skills/muse-delegate/SKILL.md) | Muse Code (`muse`) | `--disable-approval` (sandbox on); not `--yolo` [^muse] | `--read-only` (`--disable-write` + `--disable-shell`) | `--session <id>` (no headless `--resume-last`) |
| [`opencode-delegate`](skills/opencode-delegate/SKILL.md) | [OpenCode](https://opencode.ai) (`opencode`) | agent `build` (`--model` required) | `--read-only` (agent `plan`) | `--resume-last`, `--session <id>` |
| [`pi-delegate`](skills/pi-delegate/SKILL.md) | [Pi](https://github.com/earendil-works/pi-mono) (`pi`) | full local tools — no sandbox, no permission modes [^none]; project trust opt-in | `--read-only` (`read,grep,find,ls`) | `--resume-last`, `--session <id>` |
| [`omp-delegate`](skills/omp-delegate/SKILL.md) | [Oh My Pi](https://github.com/can1357/oh-my-pi) (`omp`) | `--yolo` (`tools.approvalMode: yolo`); project `.omp` extras off unless `--approve` | `--read-only` (`read,grep,glob`) | `--resume-last`, `--session <id>` |
| [`qoder-delegate`](skills/qoder-delegate/SKILL.md) | [Qoder](https://docs.qoder.com/en/cli/quick-start) (`qodercli`) | `auto` permission mode; bypass opt-in | `--permission-mode plan` | `--resume-last`, `--resume <id>` |
| [`vibe-delegate`](skills/vibe-delegate/SKILL.md) | [Mistral Vibe](https://github.com/mistralai/mistral-vibe) (`vibe`) | `accept-edits`; `--full-access` opt-in | `--plan-only` (`plan` agent) | `--resume-last`, `--session <id>` |
| [`copilot-delegate`](skills/copilot-delegate/SKILL.md) | [GitHub Copilot CLI](https://docs.github.com/copilot/how-tos/copilot-cli) (`copilot`) | `--allow-all-tools` opt-in; headless auto-deny otherwise | `--read-only` (`--mode plan`) | `--resume-last`, `--session <id>` |
| [`warp-delegate`](skills/warp-delegate/SKILL.md) | [Warp Agent CLI](https://docs.warp.dev/cli/) (`oz`) | full local tools — no sandbox, no permission modes [^none] | — [^none] | `--conversation <id>` |
| [`zcode-delegate`](skills/zcode-delegate/SKILL.md) | [Z.AI ZCode](https://zcode.z.ai) (`zcode`) [^zcode] | `--mode yolo` | `--read-only` (`plan` mode) | `--resume-last`, `--session <id>` |
[^commandcode]: Command Code's headless mode has two states and nothing between them: a `-p` run
withholds the write, edit, and shell tools, and `--yolo` (alias `--dangerously-skip-permissions`)
allows every tool anywhere the process can reach. `--permission-mode auto-accept` and `--tools-all`
do **not** lift the write gate. So an implementation run is full-trust with no path restriction —
the brief's path list is guidance, not containment. A worktree isolates the checkout, while a
container or another OS-enforced sandbox is required when writes outside the target tree are
unacceptable. `touchedFiles` is a review aid based on `git status`; it cannot show ignored files or
writes outside the repository.
[^none]: No CLI-enforced read-only mode. `touchedFiles` and the diff are what you review against, not
a guarantee: they are post-run `git status` in the workspace, so they cannot show ignored files,
reverted edits, or writes outside the repository.
[^aider]: Aider is the one implementer here that commits by default. Its `--auto-commits` and
`--dirty-commits` both default to `True`, the second of which commits your pre-existing uncommitted
work before editing. The relay always passes `--no-auto-commits` and `--no-dirty-commits`, and neither
is configurable through it.
[^grok]: `grok` cannot be prevented from writing headlessly. The relay reports a tri-state
`readOnlyViolation` tripwire for detected Git-visible changes; it does not enforce or attribute them.
[^muse]: Write `muse exec` keeps Muse's sandbox on and passes `--disable-approval` so a headless run
cannot hang on a tool-approval prompt. `--read-only` adds `--disable-write` and `--disable-shell`.
`--yolo` is not the default and is not exposed. Upstream's launcher is Darwin/Linux only. Headless
resume is `--session-id`; `muse resume` is TUI-only.
[^zcode]: ZCode ships its CLI **inside the desktop app** — there is no `zcode` on PATH, no npm
package, and the public docs cover only the GUI. The relay resolves it from
`--zcode-path`/`ZCODE_CLI`, then PATH, then the installed app bundle. Of ZCode's four documented
modes only `plan` and `yolo` work headlessly: `build` and `edit` have no permission client there, so
they block every write tool and exit 0 having changed nothing, and the relay rejects them rather
than report that as success. ZCode offers `--disallowed-tools` but no `--allowed-tools`, so
capability can be subtracted, never enumerated. Where `zcode login` fails with `OAuth response is
not valid JSON`, the key comes from `ZCODE_API_KEY` / `ANTHROPIC_API_KEY` / `ZAI_API_KEY` instead.
Each skill name links to its `SKILL.md`, which owns that implementer's prerequisites, flags, and
caveats. Building one for another CLI? [Claim it first](../../issues?q=is%3Aissue+label%3Aimplementer),
then see [CONTRIBUTING.md](CONTRIBUTING.md).
## Optional Codex waiting support
Install [codex-background](skills/codex-background/SKILL.md) alongside the selected delegate skills
when Codex is the orchestrator. It is a separate installation; codex-delegate does not install
or configure it automatically. If its MCP tools are missing, Codex runs the skill's
`scripts/bootstrap.mjs` (after telling you it will update `$CODEX_HOME/config.toml`, with a backup);
you can run it yourself with `node <skill-dir>/scripts/bootstrap.mjs`. It registers the installed
`*-delegate` skills, writes the registry and the MCP server entry, and asks you to restart Codex. One
pending MCP request captures logs locally and returns a single completion; the selected skill still
owns permissions, results, review and landing.
Claude's background Bash workflow stays unchanged. See
[configuration and recovery](skills/codex-background/references/configuration.md). Local protocol
fixtures cover the waiting boundary; they do not measure Codex Desktop token usage.
## Install
Browse first:
```bash
npx skills add amElnagdy/delegate-skills --list
```
Install the package, the setup skill, or one implementer skill:
```bash
npx skills add amElnagdy/delegate-skills
npx skills add amElnagdy/delegate-skills --skill delegate-setup
npx skills add amElnagdy/delegate-skills --skill codex-delegate
```
To pin an installation, append an existing release tag as `@vMAJOR.MINOR.PATCH`. The Skills CLI
installs by git ref, not by `metadata.version` in `SKILL.md`.
Install for a specific agent, or globally:
```bash
npx skills add amElnagdy/delegate-skills --skill codex-delegate --agent claude-code
npx skills add amElnagdy/delegate-skills --global
```
Works with any orchestrating agent the [Skills CLI](https://github.com/vercel-labs/skills) supports.
## How delegation works
Whether you choose the implementer directly or through a fleet lane, every dispatch follows the same
review-first loop:
1. **Write a brief** — self-contained task context; the implementer has no orchestrator chat history.
2. **Dispatch** it with the bundled `relay.mjs`.
3. **Wait** for completion — the relay writes a structured `result.json`.
4. **Review** the diff — re-run the project's gates yourself; pair with [guard skills](https://github.com/amElnagdy/guard-skills).
5. **Land** it — *you* commit, because committing belongs to the reviewer.
```text
Use $claude-delegate to have a separate Claude Code session implement the parser fix, then review and commit it.
Use $opencode-delegate with --lane feature to implement the billing workflow, then review and commit it.
Use $codex-delegate to run this queue of migration tasks through Codex while I review each one.
```
Every relay speaks the same `delegate-relay.result.v1` contract: `status`, `exitCode`, `signal`
(with a host-killed hint when the OOM killer ends a run), the implementer's own final report,
`touchedFiles`, and a session id where the CLI exposes one. Learn the loop once, swap the implementer
freely.
You feel it when a bounded task — a migration, a mechanical refactor, a removal sweep — comes back as
a clean diff with a structured report, and you land it after re-running the gates yourself instead of
typing it all by hand.
## What counts as an implementer skill
Four invariants hold for every `*-delegate` skill. They are also the bar for a new implementer:
- **A separate CLI edits a real working tree, and the diff is the deliverable.** Not an API wrapper,
not a gateway — an implementer whose work you can read with `git diff`.
- **The relay never commits.** Committing belongs to the reviewer, always.
- **Node built-ins only.** No dependencies, no network calls of its own, no credentials, no telemetry.
The relay launches its implementer CLI and `git`, plus the platform process launcher where a Windows
shim or a process-tree kill needs one.
- **Autonomy is stated in the CLI's own terms**, and whatever it cannot enforce is said plainly — see
the two footnotes above.
This is a loop, not a forwarder: a forwarder hands over one task and returns the output. Here you
dispatch, poll, review, and land, across one task or a queue. It stays complementary to a vendor's own
plugin or subagents — those coordinate inside one agent; this keeps the contract portable across
orchestrators, with the commit on the reviewer.
`delegate-setup` is the setup-skill exception: it discovers CLIs and writes an approved fleet map, but
never dispatches coding work.
Full checklist: [CONTRIBUTING.md](CONTRIBUTING.md).
## Requirements
- For a `*-delegate` skill, its implementer CLI authenticated as you would at the terminal. Each
implementer skill's `SKILL.md` carries its own install and login commands.
- `delegate-setup` requires no implementer CLI; it discovers whichever ones are available.
- Node 18+ and `git`.
- An orchestrating agent that can run shell commands and read files.
- Shell examples assume bash/zsh (macOS/Linux, or Git Bash/WSL on Windows).
## Trust and validation
This package is intentionally inspectable:
- All skill content is Markdown, plus small Node scripts. Each `*-delegate` skill has exactly one
`scripts/relay.mjs`. The `delegate-setup` utility ships `discover.mjs` / `config.mjs` / `lane.mjs`
(and a shared implementer table) instead of a relay — it never dispatches coding work.
- Those scripts make no network calls of their own, read or write no credentials, send no telemetry, and
have no dependencies (Node built-ins only). Relays launch an implementer CLI and `git`, plus the
platform process launcher/termination utility where a Windows shim or process-tree kill requires one.
The optional `codex-background` utility adds a local stdio MCP server around registered
existing relays, with no npm dependencies or network calls of its own. Its `bootstrap.mjs` writes
`$CODEX_HOME/config.toml` (a backup is made first) and the registry only when you or Codex run it,
and registers just the `*-delegate` skills found next to it. Its provider-independent
ownership supervisor uses a compiled subreaper on Linux and inbox PowerShell plus kernel job
APIs on Windows. Linux requires /usr/bin/cc and procfs; other hosts fail setup explicitly.
Discover may invoke installed CLIs for `--version` / model list probes (those CLIs may contact their
own services). Read the script before you run it.
- None of the relays ever commit — committing is always the orchestrator's job, after review.
**Verification status** — claims here are backed by runs, not assumptions.
True of every relay: argument handling, exit codes, `result.json` shape, supported resume mappings,
and signal reporting are verified, along with each implementer-specific guard.
Per skill — platform, CLI version, and what the run exercised:
- `aider-delegate` — Windows, `aider` 0.86.2: contract-tested against the shared smoke matrix, plus
live headless `--message-file` runs against a **stub** OpenAI-compatible endpoint on loopback. Those
runs covered: an applied edit left uncommitted, with a pre-existing dirty file still uncommitted,
proving `--no-auto-commits`/`--no-dirty-commits`; no `.gitignore` written, proving `--no-gitignore`;
a `--read-only` (`--dry-run`) run that left the target file byte-identical; an endpoint returning
401, where aider exits 0 and the relay reports `failed` with `litellm.AuthenticationError`;
`aider_unavailable`/127 writing a result file; and usage errors exiting 2 without one. Review
follow-ups were re-verified the same way: a successful run whose report says `OPENAI_API_KEY` three
times still reports `completed`; a reused `--out-dir` seeded with another run's `final.txt` and
`result.json` publishes neither; a plain exit 7 carries an `error`; and a `--read-only` run over a
modified `.aider.conf.yml` plus generated history and tags-cache warns about exactly the config
file. Not run against a hosted provider model or a real local inference server, and not run on
macOS or Linux.
- `agy-delegate` — Windows 10, native, `agy` 1.1.12: headless `--print` write run editing one briefed
file; `--read-only` `--effort high` run whose brief ordered an immediate file write, in a directory
the permission rules allowed: agy refused, wrote nothing, and `result.json` reported effort high,
`readOnly` true, `readOnlyViolation` false; argument validation for a bad `--effort` value and for
`--read-only` combined with `--dangerously-skip-permissions`, both exiting 2; resume by
`--conversation` with a delta brief. macOS, `agy` 1.0.16: headless edit run, `--print=` delivery,
absolute `--add-dir` workspace pin. `--stall-timeout` is contract-tested, live stall pending: a fake
whose log keeps growing without `streamGenerateContent` lines reports `stalled`, steady generation
calls split across writes keep a run alive, and the watchdog is off unless passed; run on native
Windows against the compiled fake.
- `claude-delegate` — macOS, `claude` 2.1.220: write run under `acceptEdits`; plan mode refusing an
edit, with the Git tripwire true on a violation and false on a clean run;
`--session`/`--resume-last` resume; `claude_unavailable`/127 and usage errors exiting 2 without a
result file; deny rules and the shell sandbox blocking `git commit`, `git push`, `git -C <dir> push`,
a nested `claude`, and a `$HOME` write.
macOS, `claude` 2.1.271: `--autocompact` argument handling only. The installed CLI accepted `auto`, `400k`, `1M`, `200`, and `200000` and rejected `auto2`, `50k`, `2m`, `99`, `1000001`, and `0` with exit 1, each observed through `claude --autocompact <value> --help`, which validates the option and exits before any model call; `claude` 2.1.220 rejects the option itself with `unknown option '--autocompact'`, which is the documented 2.1.221 floor. The relay's pass-through, resume re-pass, and `result.json` record are contract-tested against the smoke matrix; no delegated run has been dispatched with `--autocompact` set.
Windows 11, native, `claude` 2.1.274 through the npm `claude.cmd` shim, launched from Windows
PowerShell 5.1 with an `--out-dir` containing a space: write run under `acceptEdits` editing one
file and creating another, with `git status` run through the pre-approved PowerShell tool and a
`git commit` attempt denied by the relay's deny rule (`permission_denied` with
`decision_reason_type: "rule"` in the event stream, `HEAD` untouched); `--session` resume whose
delta brief referred only to "the notes file you created a moment ago" and was applied correctly;
`--read-only` plan-mode run ordered to write two files over an already-dirty tree, which wrote
nothing and reported `readOnlyViolation: false` with the dirty files byte-identical; a 45s
`--timeout` firing on a run blocked in a foreground `ping` grandchild, reporting `timeout` and
exit 1 with no `ping`, `cmd`, or `claude` process left behind; `claude_unavailable`/127 writing a
result file, and usage errors exiting 2 without one. The runs used `--model haiku`. Not run on
Windows: a directly spawned `claude.exe`, the abort path, and a read-only violation. Claude's shell
sandbox is unsupported there, so the write profile's PowerShell is not isolated.
- `cursor-delegate` — Windows, `cursor-agent` 2026.07.23-e383d2b: write run under `--force`; plan-mode
`--read-only` touching nothing; `--session <id>` resume applying a delta brief; usage errors exiting
2. A maintainer-run native macOS plan-mode smoke against the same version captured model, session,
and usage with no touched files. `--stall-timeout` is contract-tested; the live hang it targets was
observed on Linux with `cursor-agent` 2026.10.01-e373342 (two runs silent after `thinking`
`completed`, never exiting).
- `grok-delegate` — macOS, `grok` 0.2.101: streaming-json report capture, file-based brief delivery,
resume; read-only is best-effort by measurement, hence the violation flag.
Contributor-run native Windows check with `grok` 1.0.13: a no-edit dispatch using
`--trust-git-root` on a drive rejected by Git's ownership check completed with the pre-existing
untracked file reported and `readOnlyViolation: false`; that run used a real drive path whose
spelling did not diverge, and the verdict is claimed for Windows paths whose spelling does not
diverge — exact-head CI evidence for spelling-divergent temp paths lands with this repair.
Ownership, nested-directory, linked-worktree,
dirty-path, submodule, and unavailable-Git cases are covered by the `grok-git-trust` smoke module.
- `kimi-delegate` — macOS, `kimi` 0.24.0: headless `-p` edit run, stream-json parsing, and both
resume paths — the relay's `--session`/`--resume-last`, which drive Kimi's own `--session` and
`--continue`.
- `pi-delegate` — macOS: stdin brief delivery, explicit provider and model selection, JSON
session/provider/model/usage capture, and a `--read-only` run leaving a clean tree. Write,
`--session`, and `--resume-last` runs are contributor-reported.
- `omp-delegate` — contract-tested, live run pending: stdin brief delivery, `omp --mode json`
argv (`--yolo`, `--tools read,grep,glob`, `--no-extensions --no-skills --no-rules`, `--thinking`),
session header / `message_end` parsing, `--approve` omitting the project-trust flags, `--continue`
resume, assistant `stopReason: error` reported as failed, `omp_unavailable`/127, and bounded
`--version` preflight. Native Windows launch is a native `omp.exe` (no `shell:true`); that path is
contract-tested via the smoke matrix's compiled fake, not against a live Oh My Pi install.
- `qoder-delegate` — macOS, `qodercli` 1.0.47, by the contributor: Lite edit run, `accept_edits`,
explicit model and 32768-token context window, no commit.
- `commandcode-delegate` — macOS, `cmd` 1.26.0: **live edit run verified**. A relay dispatch against a
throwaway git repository had Command Code fix a remainder-dropping bug in a money-splitting function
and add three tests; the project gate was re-run independently by the orchestrator (2 tests before,
5 passing after), the diff matched the brief with no writes outside the two named files, and `HEAD`
was untouched — the relay does not commit, and the run did not either. A second dispatch with
`--session <id>` verified resume through the relay: a one-line delta brief amended exactly the comment
it named, with the session id from the first run. A `--read-only` dispatch verified the other
direction, returning `readOnlyViolation: false` on a clean tree. Also verified negatively: separate
live runs confirmed `--tools-all` and `--permission-mode auto-accept` leave the headless write gate
closed and only `--yolo` opens it.
Live running surfaced a CLI limitation the relay now handles. `cmd` ends a run with a `run_end` event
embedding the whole conversation, then exits with `process.exit`, discarding whatever is still queued
in its stdout pipe: both write runs lost their `result` line entirely (one cut ~8 KB into `run_end`,
the other losing its last ~780 events). So nothing load-bearing is read from that tail — `sessionId`
comes from `run_start`, the first line of the stream, and the report from the last `message_end` or
its streamed deltas — the event log is written in batches so the relay drains the pipe as fast as it
can, `resultLine` reports `complete`/`truncated`/`absent` so a consumer knows which fields are
trustworthy. A complete non-success result converts a zero child exit to relay exit 1, while a lost
result line falls back to the process exit code. Smoke cases pin that contract. On a long run the
report itself can land in the discarded region. The diff is the deliverable, and a thin report
means missing information, not a failed run.
Native Windows launch is contract-tested against the installed `cmdc.cmd` shape, including stdin
brief delivery and the `cmd.exe` collision guard. A live native Windows Command Code run remains
unverified; upstream recommends WSL for stable Windows use.
- `warp-delegate` — macOS, `oz` 0.2026.05.27.15.44.stable_01: **live edit run verified**. A relay
dispatch against a throwaway git repository had Warp add a function plus four assertions across
two files; both project gates were re-run independently by the orchestrator, the diff matched the
brief, and `HEAD` was untouched. Verified end to end: version preflight, launch, ndjson parsing
(`run_started` → `runId`/`runUrl`, `conversation_started` → `conversationId`), report extraction
from `{"type":"agent","text":…}` events with `agent_reasoning` excluded, `touchedFiles`, and
`status: "completed"` / exit 0. A second dispatch with `--conversation` verified resume: a delta
brief saying only "the function you just added" — never naming it — produced exactly the right
edit, with `resumed: true` and the conversation id preserved. Also observed on a prior run:
`touchedFiles: []` on a clean tree and exit 1 → `status: "failed"`. Two caveats are documented in the skill rather than fixed,
because they are Warp's behaviour and not the relay's: `finalMessage` is the agent's full
narration rather than a distinct final-message event, and `--cwd` governed shell commands while
the agent's file tool resolved bare relative paths against `$HOME`. `--no-snapshot`, `--profile`,
`--skill`, and `--mcp` are contract-tested only.
- `zcode-delegate` — Windows, `zcode` 0.16.1: read-only (`plan`) run leaving a clean tree with the
Git tripwire false; write run under `yolo` creating the briefed file and reporting it in
`touchedFiles`; `--session` resume with an attached delta brief, which recalled the earlier turn;
single-document `--json` parsing; `--version` preflight; discovery resolving the CLI from the app
bundle rather than PATH; and environment-variable auth under all three names ZCode accepts —
against an isolated home whose config carried no `apiKey`, a keyless run failed first, then
`ZAI_API_KEY`, `ZCODE_API_KEY`, and `ANTHROPIC_API_KEY` each completed the same read-only
dispatch. Contract-tested: `build`/`edit` rejection, the missing-CLI path, tolerance of the AI SDK
banner that ZCode can print on stdout ahead of the JSON (observed in direct CLI probes; exercised
in the suite by the fake), and the timeout matrix. The abort matrix is POSIX-only — Windows
delivers no catchable SIGTERM — so for this relay it first runs in CI. `zcode-delegate` is also
absent from the shared read-only tripwire scenario matrix, which runs `claude` and `grok` only —
its tripwire helpers are parity-enforced byte-identical, but no zcode-specific worktree-state run
is recorded. No macOS or Linux run is recorded.
- `codex-delegate` — macOS, `codex` codex-cli 0.150.1: fresh `workspace-write` dispatch against a
throwaway repo created the briefed file and reported it in `touchedFiles`, with `status:
"completed"` and exit 0; a `--read-only` dispatch left the tree clean (`touchedFiles: []`) and
returned a `threadId`; a follow-up `--session <id>` resume against that thread was asked to recall
a word from the first turn without it being named again — the delta brief said only "the word you
picked a moment ago" — and it answered correctly, confirming the resumed turn saw prior context, not
just that the id was accepted. Contract-tested: argument validation,
bounded version preflight, missing binary, result parsing, and whole-process-tree timeout/abort
cleanup. Windows 11, `codex` codex-cli 0.153.4, native (Git Bash launch, no `pwsh` installed):
a `workspace-write` dispatch against a throwaway repo created the briefed file and reported it
in `touchedFiles` with `status: "completed"`, exit 0, and a `threadId`; the brief had Codex run
shell commands and report its shell, which came back as System32 `powershell.exe` 5.1 with zero
`WindowsApps` entries on its `PATH` while the parent process carried one — the relay's PATH
filter reaching the sandbox, not just the child's argv. The denial the filter exists for was
reproduced through the unpatched relay on the same machine: a `workspace-write` dispatch asked
Codex to run `winget`, which lives only under `WindowsApps`, and both the PATH lookup and the
absolute path failed inside the sandbox with "The file cannot be accessed by the system" while
the same binary ran normally outside it. No Store `pwsh` is installed here, so the shell-launch
form of that denial (`0xC0070005` on every command) rests on the report in issue #116. No Linux
run is recorded.
- `muse-delegate` — macOS, Muse Code 1.4.7: a fresh write dispatch against a throwaway repo
created only the briefed file (`touchedFiles` reported exactly it, `status: "completed"`, exit 0,
a `sessionId` captured); a `--session <id>` continuation in that same session appended the briefed
line; a `--read-only` dispatch answered from the tree and left it byte-identical; and an
`--provider echo` dispatch completed end to end. Contract-tested: argv exactness (`muse exec`
with `--disable-approval` / `--disable-write` / `--disable-shell`, no `--yolo`, `--session-id`,
`--resume-last` rejected), exit 0 without `run.terminal.completed` reported `failed`, bounded
version preflight, missing binary, the bounded `stderrTail` over a flooded stderr, and
whole-process-tree timeout/abort cleanup. Upstream's launcher is UNIX-only. No Windows or Linux
run is recorded.
- `opencode-delegate`, `vibe-delegate` — contract-tested only: argument validation, bounded version
preflight, missing binary, result parsing, and whole-process-tree timeout/abort cleanup. No
end-to-end run is recorded here.
- `cline-delegate` — macOS, `cline` 3.0.52: current-binary unauthenticated plan probe reached
`run_start` with the fixed positional instruction plus the real brief on stdin, accepted a
provider-local model id, parsed the failing `run_result`, and left the tree clean. Contract-tested:
plan mode forcing `--auto-approve false`, the unsafe true conflict, argument validation, nullable
`sessionId`/`finalPath`, bounded version preflight, missing binary, result parsing, and whole-process-tree
timeout/abort cleanup. The contributor also reported a native Windows 3.0.51 edit run against the
earlier positional-brief commit; that does not verify this exact stdin-based head on Windows.
- `copilot-delegate` — Windows, `copilot` 1.0.78: `--read-only` plan-mode run completed with a clean
tree and captured session id; `--allow-all-tools` edit run created the requested file; the headless
auto-deny path was exercised live (denial detected from the data-wrapped event shape, run reported
failed with the `--allow-all-tools` hint); `--session <id>` and `--resume-last` resume runs executed
their delta briefs via the directive-wrapped `-p @<file>` prompt. Contract-tested: argv exactness
(including the resume directive), denial shape, `--read-only`/`--allow-all-tools` conflict
validation, bounded version preflight, missing binary, result parsing, and whole-process-tree
timeout/abort cleanup.
- `delegate-setup` — contract-tested: discover JSON shape, config validate/write/load, whole-lane
project overlay, global write without creating `.delegate/`, and `--lane` resolve / wrong-skill /
flag-override against relays. The smoke suite runs live discovery against installed CLIs
(versions vary by machine). Native Windows discover smoke not yet claimed.
Not yet verified: native Windows launches for a directly spawned `claude.exe` (the npm `claude.cmd`
shim path has a run above), exact-head `cline`, `grok`, `kimi`, `muse`,
`pi`, `qoder`, `vibe`, and `omp` (`codex`/`opencode`/`grok`/`commandcode` have contract-tested `.cmd` shim handling;
Cursor serializes a pre-joined, quoted command; Qoder and Vibe target their documented native executables).
Claude's own shell sandbox is unsupported on native Windows regardless of launch mechanics, upstream
Vibe officially targets UNIX, and Muse's upstream launcher is UNIX-only. A native Linux `cursor-agent` run is unverified. The full delegate →
review → commit loop is designed for and run on Claude Code; other orchestrators (Cursor, …) are
designed-for but unproven.
## Repository shape
Implementer skills share one shape; the utilities have different ones:
```text
skills/
├── <name>-delegate/
│ ├── SKILL.md
│ ├── scripts/relay.mjs
│ └── references/
│ ├── writing-the-brief.md
│ ├── dispatch-and-poll.md
│ ├── review-and-land.md
│ └── multi-task-queues.md
├── codex-background/
│ ├── SKILL.md
│ ├── scripts/
│ │ ├── bootstrap.mjs
│ │ ├── bootstrap.test.mjs
│ │ ├── server.mjs
│ │ ├── server.test.mjs
│ │ ├── supervisor.mjs
│ │ ├── linux-owner.c
│ │ ├── windows-job.ps1
│ │ └── windows-job.cs
│ └── references/configuration.md
└── delegate-setup/
├── SKILL.md
├── scripts/
│ ├── discover.mjs
│ ├── config.mjs
│ ├── lane.mjs
│ └── implementers.mjs
└── references/
├── schema.md
└── setup-dialogue.md
```
Adding an implementer is a new directory plus two lines here: a table row, and a verification line once
a run backs it.
Contributing? House rules, the controlled vocabulary, and the pre-publish checklist live in
[AGENTS.md](AGENTS.md) — read it before opening a pull request, and point your agent at it too.
## License
MIT — see [LICENSE](LICENSE).