delegate
Route one task file to the cheapest capable model lane, validate the output, and return a single verdict line.
Skill by Sovereign-Communication
Chalk Labs
Need help deploying delegate?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/sovereign-communication-delegate
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: productivity
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/delegate
What's inside
- Skill name
delegate- Description
- Route one task file to the cheapest capable model lane, validate the output, and return a single verdict line. Use when work is mechanical, scoped, or parallel and does not require a human judgement call. Keeps orchestrator context clean - worker output goes to a file, only PASS/BLOCKED comes back.
- License
- Unlicense
- Instructions
- 91 lines in SKILL.md · view on GitHub
Agent Skills spec issues
- files: referenced but missing: scripts/lanes.json
Files (1)
- SKILL.md
SKILL.md (first lines)
# delegate
Send work out. Keep your context clean. Get back one line.
```bash
python scripts/delegate.py --task <file> --tier <tier> [--files a.rs b.rs] [--mode diff|full]
```
Exit `0` PASS, `3` no usable output after retries, `4` BLOCKED (no capable lane
or task needs a shell), `5` task file missing.
## Choosing a tier
The tier says what the work *needs*, not who does it. Routing is derived from
that plus the live roster.
| Tier | Use when | Notes |
|---|---|---|
| `micro` | Mechanical, one file, no design judgement | Sub-second lanes exist |
| `scoped` | A real diff against a written spec | The common case |
| `reasoning` | Diagnosis or design where being confidently wrong is the risk | Heavier, slower lanes |
| `long-context` | Many files must be in one prompt | 512k and 1M ctx lanes |
| `shell` | Must actually RUN something (`gh`, `cargo`, `adb`, `gradlew`) | HTTP lanes cannot; goes to agy |
| `verdict` | Go/no-go with human consequences | Never delegate. That is your job. |
`delegate.py` refuses to send a shell-needing task to an HTTP lane and tells you
to re-dispatch with `--tier shell`. Three lanes each burned a call discovering
this by hand before the guard existed.
## Do not delegate
- One-liners. Dispatch overhead exceeds the task.
- Bounded 1-5 tool-call diagnostics. Cheaper inline.
- Anything whose failure mode is "confidently wrong and merged anyway."
- Verdicts. A worker cannot verify itself; that is the only reason your tier exists.
## Writing a task file that works
The task file IS the prompt. Every failure traced so far came from an
underspecified one.
1. **State the output contract.** `delegate.py` appends one, but say it again in
your own words: a diff, a whole file, a table.
2. **Give it an out.** Tell it to reply `BLOCKED: <reason>` if it cannot comply.
A model that says BLOCKED is worth more than one that invents a plausible
answer, and the validator treats BLOCKED as a real result.
3. **Pin the scope.** Name the files it may touch. Say what NOT to reformat.
4. **Split diagnosis from repair.** One task extracts the error, a second fixes
it and is forbidden from starting without that output. Fixing from a
hypothesis is how repeated attempts fail.
5. **Never let it self-certify.** "Run the tests and confirm they pass" returns a
claim. Require the command output, and have someone else run it.
## Reading the result
`PASS` writes to `tmp/delegate/<task>__<lane>.md` with lane, model, and latency
in the header. **Read the diff before you trust it.** A PASS means the output
was well-formed, not that it was correct.
`BLOCKED` lists every lane tried and why. Escalate deliberately, in this order,
stopping as soon as one works:
1. Re-dispatch with a **better task file** - most blocks are spec defects, not
lane defects.
2. `--tier long-context` if it plausibly ran out of room.
3. `agy-gemini` - the only free lane with a shell, so it can verify itself.
Always `--add-dir <repo>` and always pin `--model` to an exact name from
`agy models`. On timeout use `--continue`, never re-dispatch fresh.
4. `agy-claude` - spends Anthropic quota from the same pool as your session.
Only after free lanes failed twice on the same task with different contracts.
5. Native. You. Last.
## Keeping the roster honest
```bash
python scripts/delegate.py --list-lanes # current capacity + staleness date
python scripts/lane_probe.py # re-measure; run weekly or after any 401/404
```
Lanes die without warning. Between 2026-08-04 and 2026-08-15, Qwen and DashScope
both went to 401 and OpenRouter retired four `:free` tiers. `scripts/lanes.json`
carries an explicit expiry for that reason - if it is stale, re-probe rather than
trusting it.
**The failure mode that will bite you:** free reasoning models spend their whole
token budget on hidden reasoning and return empty content. That is not a refusal
and not a broken lane. `delegate.py` sends `reasoning:{effort:low}` to OpenRouter
automatically, and must never send it to Google, NVIDIA, Cerebras or Groq, which
reject it. See `traps` in `scripts/lanes.json`.From Sovereign-Communication/SCMessenger, Unlicense.
More from Sovereign-Communication/SCMessenger
- 0.609score
ceo
Sovereign-Communication
Resume the SCMessenger CEO seat - load state, re-derive live node and repo state, audit the CTO seat's checkpoints against the tracked package, and maintain…
SkillAFreshness: A★ 4Claude Code +9 - 0.600score
dreaming
Sovereign-Communication
Run a dreaming lessons-learned distillation pass.
SkillAFreshness: A★ 4Claude Code +9 - 0.569score
bod
Sovereign-Communication
Convene the SCMessenger Board of Directors - evaluate key architectural, cryptographic, and philosophical proposals using a 5-judge panel with…
SkillBFreshness: B★ 4Claude Code +9 - 0.566score
onboard
Sovereign-Communication
Seat-onboarding checklist for SCMessenger - load when taking the CTO or orchestrator seat in this repo, before any dispatch, merge, or verdict.
SkillBFreshness: B★ 4Claude Code +9 - 0.565score
cto
Sovereign-Communication
Resume the SCMessenger CTO seat - load state, re-derive live three-node state, follow the tracked V040 three-node BLE package, checkpoint at each stage, and…
SkillBFreshness: B★ 4Claude Code +9 - 0.561score
build-verify
Sovereign-Communication
Run the SCMessenger build verification gates (cargo build/check/clippy/fmt/test-compile, and optionally Android/WASM).
SkillBFreshness: B★ 4Claude Code +9 - 0.554score
orchestrate
Sovereign-Communication
Portable SCMessenger Orchestration Control Plane v2 discovery bootstrap.
SkillBFreshness: B★ 4Claude Code +9 - 0.554score
orchestrate
Sovereign-Communication
SCMessenger Orchestration Control Plane v2 bootstrap for Bob.
SkillBFreshness: B★ 4Claude Code +9 - 0.513score
gemini-orchestrator
Sovereign-Communication
# Gemini 3.5 Flash Orchestrator — Foreign-worker coordination for backlog continuity
SkillCFreshness: C★ 4Claude Code +9 - 0.512score
scmorc
Sovereign-Communication
SCMessenger Headless Orchestrator — per-task claude.exe workers with tuned model + effort, native Anthropic subscription only
SkillCFreshness: C★ 4Claude Code +9 - 0.511score
docs-sync
Sovereign-Communication
Run the repo's documentation sync check (scripts/docs_sync_check.sh).
SkillCFreshness: C★ 4Claude Code +9 - 0.511score
docs-sync
Sovereign-Communication
Run the repo's documentation sync check (scripts/docs_sync_check.sh).
SkillCFreshness: C★ 4Claude Code +9
Signals
- Stars
- 4
- Forks
- 2
- Open issues
- 83
- Last commit (this folder)
- Aug 15, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Unlicense
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from Sovereign-Communication
- 0.609score
ceo
Sovereign-Communication
Resume the SCMessenger CEO seat - load state, re-derive live node and repo state, audit the CTO seat's checkpoints against the tracked package, and maintain…
SkillAFreshness: A★ 4Claude Code +9 - 0.600score
dreaming
Sovereign-Communication
Run a dreaming lessons-learned distillation pass.
SkillAFreshness: A★ 4Claude Code +9 - 0.569score
bod
Sovereign-Communication
Convene the SCMessenger Board of Directors - evaluate key architectural, cryptographic, and philosophical proposals using a 5-judge panel with…
SkillBFreshness: B★ 4Claude Code +9 - 0.566score
onboard
Sovereign-Communication
Seat-onboarding checklist for SCMessenger - load when taking the CTO or orchestrator seat in this repo, before any dispatch, merge, or verdict.
SkillBFreshness: B★ 4Claude Code +9 - 0.565score
cto
Sovereign-Communication
Resume the SCMessenger CTO seat - load state, re-derive live three-node state, follow the tracked V040 three-node BLE package, checkpoint at each stage, and…
SkillBFreshness: B★ 4Claude Code +9 - 0.561score
build-verify
Sovereign-Communication
Run the SCMessenger build verification gates (cargo build/check/clippy/fmt/test-compile, and optionally Android/WASM).
SkillBFreshness: B★ 4Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get delegate running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
delegate skill. Agent Mart. https://agentmart.dev/skills/sovereign-communication-delegate (last updated 2026-10-24)