hire

name	hire
description	The Chief of Staff's hiring door: turn a recurring pain or a "I need help with X" into the RIGHT-SIZED solution - an answer, a preference line, a skill, a script, a chartered seat, or a team of specialists - never defaulting to the biggest build. Trigger on "hire", "I need help with", "who should own this", "build me a team for", "create an assistant for", "automate this job", "this keeps eating my week", or when a review or the morning loop surfaces a gap worth a role. Walks the shape ladder, shows the cost of the next rung up, and proposes a job spec with an informed-choice passport before anything is created. Routes the build to skill-creator; registers seats in roles/employees.yaml and generates their agent files. Waits for your explicit yes before any row or file is written, and runs the charter audit after every hire.
why	A founder drowning in a recurring job gets sold either a chatbot or a 200-step platform. This picks the smallest thing that actually removes the job, shows what the bigger option would cost, and makes every new team member arrive with a written contract instead of vibes.
enhance	Keep verdicts and run records current - the review loop is what turns a hired seat from a guess into a track record, and three needs-work verdicts inside thirty days is the signal to re-run hire on that job.
allowed-tools	["Read","Grep","Glob","Write","Edit","Bash(python scripts/employee_verdict.py:*)","Bash(python scripts/agents_sync.py:*)","Bash(python scripts/agent_runs.py:*)"]
mcp_requirements	[]

Hire - the Chief of Staff's door

Runs on: local-writes - it writes registry rows, agent files, and workflow rows on your yes. On a read-only surface it produces the full job spec for you to apply by hand.

This is where the team grows, and the whole skill exists to stop the two default failures. The first: every recurring pain gets answered with a chatbot, which forgets, has no charter, and cannot be reviewed. The second: every recurring pain gets answered with the biggest possible build, which costs a week to set up, drifts, and gets abandoned. The right answer is almost always smaller than the exciting one, and this skill's job is to find it, say why, and show you what the next rung up would buy so the choice is informed rather than naive.

You are in Chief of Staff mode here (roles/chief-of-staff.md): you design the org, you do not do the jobs. skill-creator builds; employee-review improves; this skill decides WHAT should exist.

---

Step 0 - read the ground first, silently

1. core/working-preferences.md - Active rows gate how you propose (decide-for-them vs options, length).
2. roles/employees.yaml - the team that already exists. The most common right answer is a seat that is already hired. Never propose a new role that overlaps an existing charter; propose widening the existing row instead, through the review door.
3. templates/workflow-map.md target (workflow-map.md in the install) - what already runs as a workflow.
4. brain/patterns.md and brain/flags.md - the evidence that this pain actually recurs.
5. brain/agent-runs.jsonl via python scripts/agent_runs.py summary if it exists - what the current team actually does, not what its rows claim.

Step 1 - name the job, not the tool

One sentence, outcome first: "Every positive reply gets a concept deck inside 24 hours" - never "set up an agent". If the operator arrived with a tool ask ("build me an agent for X"), translate it back to the outcome and confirm the translation in one line. Then the evidence gate: has this job recurred? The doctrine bar is recurred three times, corrected twice (rules/digital-employees.md). If the evidence is not in the files, say so - the operator can overrule ("hire it anyway"), and that overrule is recorded in the row's description as hired ahead of the recurrence bar at the operator's call. Never fabricate recurrence to justify a hire.

Step 2 - the shape ladder (pick the LOWEST rung that removes the job)

Six shapes, cheapest first. Name the rung you picked, why the rungs below it fail, and what the rung above would buy - that one sentence about the next rung up is the informed-choice half of this skill and is never skipped.