Mission init

Skill hashorva/purrfect-harness/.agents/skills/mission-init

Portable orchestrator/worker harness kit for AI-assisted development

Install

npx -y skills add hashorva/purrfect-harness --skill mission-init

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

17 days oldThe repository was created 17 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Bootstrap an orchestrated mission: create a dated folder under docs/missions/ with GOAL.md, features.json, PROGRESS.md, and the first WORKER_TASK from the repo's templates. MUST be used whenever the user says "start a mission", "new goal", "bootstrap the loop", "let's run the orchestration on X", or asks to begin orchestrated work of any kind. Also use to re-scope an existing mission. Produces files mechanically from docs/missions/templates/ — never freestyle the formats. Runs fleet-inventory at GATE 0 and always shows the seat→model map for human confirm/swap. Default chair is Opus (Cursor intake plans briefly; Opus weighs and orchestrates) unless the human names otherwise.

SKILL.md

5.2 KB, as published. Nobody here has run it

Mission init (orchestrator only)

Step 0 — Preconditions

No active mission: run bash scripts/active-mission.sh. If it prints a path, STOP and ask the human whether to finish (status: done), keep waiting on a gate, or re-scope. Never start a second in-progress / awaiting-gate mission.
Also refuse leftover root GOAL.md / features.json (pre-2.0 layout) — move or delete them first.
Read AGENTS.md (hard rules constrain every feature) and skim docs/ filenames so features reference real docs.
Load docs/missions/MODEL_ROUTING.md once for this GATE 0 decision.

Chair resolution (before writing files)

Default brain = Opus. Exceptions (human must say, or session already is):

Signal	Chair
Already on Opus (Claude Code or Cursor UI)	This session
“You are the orchestrator” / “no handoff”	Current model
“No orchestration — just execute”	STOP mission-init; implement in chat
“Orchestrator = Fable”	Fable (named only — never auto-escalate)
“Orchestrator = Codex highest” / Sol	Inventory `brain_codex` / family `sol`

Cursor intake: if a brief plan exists in a Cursor (often Grok) chat, Opus reads it, weighs whether the split looks balanced, then continues mission-init. Do not treat the intake chat as the orchestrator unless named above.

Step 0b — Local CLI inventory (mandatory)

command -v claude; command -v codex; command -v agent; command -v agy
bash scripts/fleet-inventory.sh

Always present the printed seat→model map at GATE 0. Human may remap any seat (economy ↔ premium, or brain) before approval. Copy approved binds into GOAL Fleet. Missing a CLI → strike that row; do not invent invocations.

Step 1 — Extract the mission (interview only if needed)

From the user's paragraph (and any Cursor intake brief), determine: outcome, forcing function, out-of-scope, human gates, chair (above), fleet tier (economy | premium | mixed). Slug: YYYYMMDD-short-kebab-name. At most 3 clarifying questions. Do not write files until the outcome is unambiguous.

Step 2 — Write the mission folder FROM TEMPLATES

SLUG="{{YYYYMMDD-short-name}}"
MISSION="docs/missions/$SLUG"
mkdir -p "$MISSION/tasks"
cp docs/missions/templates/GOAL.template.md "$MISSION/GOAL.md"
cp docs/missions/templates/features.template.json "$MISSION/features.json"
cp docs/missions/templates/PROGRESS.template.md "$MISSION/PROGRESS.md"
cp docs/missions/templates/WORKER_TASK.template.md "$MISSION/tasks/T-001.md"

Fill placeholders. Set status: draft. Paste inventory binds into Fleet. Validate: python3 -c "import json; json.load(open('$MISSION/features.json'))". Only create tasks/T-001.md at init (later tasks after reviews).

Step 3 — GATE 0 (mandatory stop)

Brain CLI proof (mandatory unless human names current-chat chair):

bash scripts/fleet-inventory.sh
bash scripts/spawn-brain.sh "$MISSION" mission-init   # real claude --model opus
# If human said "you are the orchestrator" / already on Opus in this chat:
#   bash scripts/spawn-brain.sh "$MISSION" waiver --reason "already on Opus in Cursor"
bash scripts/verify-mission.sh "$MISSION"             # must pass before greenlight

PROGRESS prose is not proof. Greenlight = .tasks/receipts/brain-*.json with exit: 0 and cli: claude (or documented waiver).

Present:

Feature list (id + description + priority)
Full fleet inventory map (seat / CLI / model / family) — confirm or swap
Resolved brain + concrete model + brain receipt path
T-001 worker + tier (premium Grok vs economy Composer for Cursor, etc.)
Assumptions

DO NOT dispatch until the human approves the features and the bind map and verify-mission.sh is green for brain.

On approval:

Append Human — GATE 0 PROGRESS entry (include approved binds + receipt path).
Set GOAL status: approved → in-progress on first spawn.
Follow .agents/skills/dispatch-worker/SKILL.md with scripts/spawn-worker.sh … --tier … --feature F00X.

Never

Never invent file formats — templates only.
Never write features AGENTS.md forbids.
Never auto-escalate to Fable.
Never put mission files at repo root.
Never skip showing the inventory map at GATE 0.
Never flip passes: true or status: done without bash scripts/verify-mission.sh green.
Never claim Claude/Cursor CLI ran without a matching .tasks/receipts/*.json.
Never implement allow-listed worker files in the orchestrator chat when spawn-worker is available.