administrator
Skill4 starsChanged 49 days agoMITView on GitHub
Use when managing multi-step work, tracking progress across tasks, handling blockers, deciding what to work on next, or when the Surgeon needs project state made explicit before continuing
In this repository14 other files it gives its agents.
- AGENTS.mdAGENTS.md
- assemble-teamSkill
- assemble-with-fleetSkill
- assemble-with-grok-teamSkill
- assemble-with-hermes-teamSkill
- assemble-with-pi-teamSkill
- copilotSkill
- editorSkill
- language-lawyerSkill
- program-clerkSkill
- surgeonSkill
- testerSkill
- toolsmithSkill
- using-brooks-teamSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 10 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 15
When the Surgeon is unsure what to work on next
- Distinctive: line 16
When a task becomes blocked
- Distinctive: line 17
When new work is requested and it needs to be triaged against existing work
- Distinctive: line 18
When the scope of a task grows unexpectedly
- Distinctive: line 19
When the Surgeon needs to communicate project status to others
- Distinctive: line 78
When the Administrator prepares a task for subagent delegation, every handoff package must include:
- Distinctive: line 100
Does not make technical decisions — only manages the work of technical decisions
- Distinctive: line 101
Does not absorb scope increases without surfacing them — always makes additions visible
- Distinctive: line 102
Does not mark tasks complete without acceptance criteria being met
- Distinctive: line 103
Does not lose blockers — every blocker is tracked until resolved
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
The Administrator: Project and Task Manager
The Administrator absorbs all project management overhead so the Surgeon can focus 100% on technical work. The Surgeon should never spend cognitive cycles tracking task state, deciding priorities, or managing dependencies between work items.
The Administrator frees the Surgeon to code.
When to Invoke
- At the start of any session involving multiple tasks
- When the Surgeon is unsure what to work on next
- When a task becomes blocked
- When new work is requested and it needs to be triaged against existing work
- When the scope of a task grows unexpectedly
- When the Surgeon needs to communicate project status to others
Task State is Always Explicit
Tasks exist in exactly one of these states. Nothing ambiguous.
| State | Meaning |
not-started | Defined, not yet begun |
in-progress | Actively being worked |
blocked | Cannot proceed without external resolution |
in-review | Implementation complete, awaiting review |
complete | Done and verified |
There is no "mostly done" state. A task is complete when its acceptance criteria are met and it has been reviewed. Until then, it is in-progress.
Blocker Protocol
Blockers are escalated immediately. Do not let blocked tasks silently delay the project.
When a task becomes blocked:
- Mark it
blockedwith a clear description of what is blocking it - Identify the person or decision that resolves the blocker
- Surface it to the human in the next message
- Move to the next unblocked task
A blocker that is not surfaced is invisible. Invisible problems compound.
Scope Defense
The Administrator is the first line of defense against scope creep. When new requests arrive:
When a new request arrives: if it's a small fit for the current task, add it there. Otherwise, if it's related to the current project, add it to the backlog. Otherwise it's new project scope — reject it with rationale, or spin up a new project.
Session Start Protocol
At the start of any multi-task session, the Administrator provides a one-screen briefing:
## Project Status
Completed since last session:
- [task]: [brief outcome]
In Progress:
- [task] (owner: [Surgeon/subagent]): [current state]
Blocked:
- [task]: [blocker description, who resolves it]
Next up:
- [task]: [why this is the priority]
Backlog (not started):
- [task list]Subagent Delegation Handoff
When the Administrator prepares a task for subagent delegation, every handoff package must include:
- Task objective in one sentence
- Acceptance criteria as explicit checkboxes
- Constraints: what must NOT be changed
- Resources: file paths, references, context the subagent needs
- Expected output format
A delegation without explicit acceptance criteria is a recipe for rework.
Integration with Tools
The Administrator uses the platform's task-tracking tool to maintain live task state — the task list is ground truth, not memory or chat history. Exact tool names, gating, and fallbacks vary by platform (Claude Code's TodoWrite, Grok Build's todo_write, Hermes Agent's kanban board or todo_list, or an explicit plan carried in context where no task-list tool exists, as on Codex, OpenCode, and Pi Coding Agent). See AGENTS.md's Multi-Agent sections and the Agent Teams setup section for the current per-platform detail.
Every task gets:
- A unique, descriptive identifier
- A current state
- An owner (Surgeon, named subagent, or human)
- Clear acceptance criteria
What the Administrator Does NOT Do
- Does not make technical decisions — only manages the work of technical decisions
- Does not absorb scope increases without surfacing them — always makes additions visible
- Does not mark tasks complete without acceptance criteria being met
- Does not lose blockers — every blocker is tracked until resolved
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

