Methods / 7 min
A longer skill collection is not the same thing as a more capable agent.
Proposed worksheet · Method notes updated 2026-11-11 · No independent effectiveness test
The decision
Add a skill only when its task fit, authority boundaries and maintenance burden are understood.
Use this when
You are considering a reusable instruction package or updating one already used by an agent.
Skip this when
The current workflow already meets the task requirements with no repeated failure. Keep the free, competent baseline in the comparison.
Work through the checks
Name the job
Describe the exact problem the skill should solve and when it should not run. Read its instructions and referenced scripts. Inspect the requested tools, data, network access and side effects.
Check compatibility and provenance
Record the source, version, license and supported host. A valid SKILL.md name and description can help discovery but do not prove safety or quality. Treat scripts as code, not prose.
Compare against the existing workflow
Use comparable tasks, the same tools and permissions, and clear acceptance tests. Do not weaken the baseline so the new package looks useful. Include failure and unavailable-tool cases.
Keep rollback small
Preserve the prior version and a way to disable the new skill. Stage authorized changes in an isolated environment. Keep permission changes separate from content updates.
What would count as success?
- The skill improves a defined outcome or removes demonstrated recurring work. Evidence for check 1
- Quality and required controls are preserved. Evidence for check 2
- The previous working version remains recoverable. Evidence for check 3
Evidence to bring back
Proposed evidence plan, not observed results or extra mandatory process. Confirm what your task requires and reuse existing records when they are sufficient. Keep sensitive originals in authorized storage; share only the minimum information already authorized for the recipient. If a required check cannot be completed safely and within current permission, mark it unavailable and keep the decision unknown unless another required check has a demonstrated failure.
The skill improves a defined outcome or removes demonstrated recurring work.
Proposed evidence for acceptance check 1. No result has been observed here.
Bring back: Keep paired task results against the same competent baseline, with the defined outcome, recurring problem and setup/review burden visible. Record a keep or skip decision when no useful gain is demonstrated.
Not enough: Popularity, a longer output or a reduced tool count without an accepted-result comparison.
Quality and required controls are preserved.
Proposed evidence for acceptance check 2. No result has been observed here.
Bring back: Keep the acceptance results and required permission/failure-path checks for both workflows, including failures and unavailable cases. Changes to permissions need their own approval.
Not enough: A faster happy path that omits the baseline’s quality checks or weakens a required control.
The previous working version remains recoverable.
Proposed evidence for acceptance check 3. No result has been observed here.
Bring back: Keep the prior version’s identity, authorized storage location and a bounded restore check of that version in the intended test environment. A disable check demonstrates only that the new skill can be stopped. State what recovery was not exercised.
Not enough: A promised rollback or an untested backup pointer described as demonstrated recovery.
Stop or recover
Do not install when provenance, required access or effects are unclear. Stop if a proposed update silently expands permissions or makes unsupported compatibility claims.
What this does not establish
This worksheet does not install, certify or endorse any skill. A schema check is not an effectiveness test.
Reference links
These links leave the saved pack when selected. Check whether dated guidance still applies before relying on it.
- Agent Skills specification (Online source)
The skill-package format and metadata contract. Format compliance does not establish safety, quality or usefulness.
Source review date: 2026-11-10
- Anthropic: building effective agents (Online source)
Engineering guidance on fitting workflow complexity to task needs. Not a controlled evaluation of this method.
Source review date: 2026-11-10
Keep your own evidence
The draft already identifies this method and its proposed requirements. Confirm those requirements against your task, then supply only the scope and observations you are authorized to record. No evidence or permission is prefilled. Reuse an existing suitable record if you have one.
- Open this guide’s Markdown draft
- Open this guide’s JSON draft
- Inspect the guide’s source hash and method identity
Your browser may show or download text files. Edit a separate working copy in your usual editor. This reader does not accept or save your answers.
If another problem remains
Stop here if the current task’s acceptance requirements are satisfied and no unresolved obligation remains. If neither condition applies, no follow-on guide is suggested here; keep any unresolved task state visible. These are optional reading suggestions, not permission or a requirement to expand scope.
Only if: A candidate fits a documented problem, but its benefit over the current workflow is still unmeasured.
Measure improvement in accepted resultsDefine a bounded comparison before deciding to adopt it.
Only if: An approved trial needs its inputs, prior version and results preserved for review or recovery.
Make the handoff survive the sessionKeep the exact versions recoverable without treating a saved package as launch approval.