Skip to content

arjunprabhulal/agent-skills / skills/agent-lifecycle/agent-scoping

agent-scoping

Verified

Turns a vague agent idea into a reviewable agent spec — the jobs it does, the boundaries it must not cross, and the criteria that decide whether it works. Use this whenever the user talks about building an agent, an assistant, a bot, or an automation, even if they only describe the idea informally and never ask for a spec — a request like "build an agent that handles X" with no written scope is exactly the trigger. Also use it when an existing agent's purpose has drifted and needs restating. Don't use it for choosing tools, models, or architecture, or for writing eval cases, though it produces the raw material for both.

FileKindSize
SKILL.mdskill md5.6 KB
references/agent-spec-template.mdmarkdown2.4 KB
references/scoping-questions.mdmarkdown3.5 KB

references/scoping-questions.md

View on GitHub
# Scoping question bank

Use these when the four core questions in `SKILL.md` do not fit the situation. Pick the
section matching the agent type, then take the three or four questions that are actually
uncertain. Asking all of them is an interrogation, not a scoping session.

## By agent type

### Assistive — a human is in the loop on every result

The human is the safety net, which changes what matters. Speed and legibility beat autonomy.

- Can the user tell a good answer from a bad one without extra work? If not, the agent needs
  to show its evidence, not just its conclusion.
- What does the user do with the output next — paste it, approve it, act on it?
- What is the cost of a plausible-but-wrong answer that gets accepted?
- How much waiting will the user tolerate before doing it manually instead?

### Autonomous — the agent acts without review

Every action is a commitment. Scope tightly and expand later.

- Which actions are reversible, and which are not? Enumerate the irreversible ones by name.
- What is the blast radius of the worst single action? Of a loop of them?
- What signals a run has gone wrong, and who sees that signal?
- What is the kill switch, and who can pull it at 3am?
- Is there a dry-run mode, and will it be trusted enough to be used?

### Batch — runs over a queue or on a schedule

Volume converts a small error rate into a large absolute number.

- What is the volume per run, and what error rate is tolerable at that volume?
- What happens to items the agent cannot handle — dropped, queued, escalated?
- Is partial completion acceptable, or is it all-or-nothing?
- Who reads the output, and how soon after the run?
- What is the per-run cost ceiling?

## Cross-cutting

### Baseline

- What happens today without this agent? Get specifics: who does it, how long it takes, how
  often it goes wrong.
- If the answer is "nothing, this is new" — is the real problem that nobody wanted it done?
- What did the previous attempt look like, and why did it stop?

### Data

- What does the agent need to read, and does it already have access?
- Is any of it sensitive, regulated, or subject to residency rules?
- Is the data reliable enough to act on? An agent on top of a stale source inherits the
  staleness and adds confidence to it.

### Trust

- Who has to trust this agent for it to be adopted, and what would earn that trust?
- What happens the first time it is confidently wrong in front of that person?
- Is there a path from assistive to autonomous, or is the trust ceiling permanent?

### Scale of ambition

- What is the smallest version worth shipping? Push for a genuinely small answer.
- Which job, if it worked reliably, would justify the whole project on its own?
- What would make you shut this down after three months?

## Answers that mean "stop and reconsider"

Some responses are signals that the scoping problem is upstream of the agent:

- **"It should handle anything a user asks."** No boundary exists yet. Do not proceed to
  design; produce a jobs list first and get it rejected a few times.
- **"We'll know it works when it feels right."** No success criteria. Offer to draft
  criteria and have them corrected — people find it easier to react than to specify.
- **"Same as <other product>, but for us."** The comparison is doing all the work. Ask which
  three of that product's jobs actually matter here.
- **"It just needs to be an agent."** The agent is the requirement, not the outcome. Find
  the outcome, or note plainly that a simpler tool would do the job.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall