tzachbon/srulik-toolkit / plugins/srulik-toolkit/skills/skill-creator
skill-creator
UnverifiedCreate or edit a skill, improve its instructions or triggers, evaluate it on real tasks, or benchmark a revision. Every skill creation or edit uses create-plan and implement-plan, including small changes; evaluation compares actual outputs and includes human review.
Skill creator
Adapted from Anthropic's skill-creator; modified for portable execution and Srulik's mandatory planning, implementation, and review handoffs. Original terms and copyright are retained in LICENSE.txt.
Bind the request
Identify the target skill, requested outcome, scope, output location, existing instructions, and observable completion evidence. Inspect the target before editing; preserve an existing name and invocation policy unless the request changes them. Skill creation, editing, trigger optimization, and feedback-driven revisions all count as authoring.
Read authoring guidance for authoring and evaluation guidance for real test runs. Select the matching host mapping. These mappings name the supported hosts and concrete tools. On an undocumented host, show the capability mapping, label it untested, and obtain confirmation before proceeding. Missing capabilities remain explicit limitations.
Plan every creation or edit
Invoke create-plan for every skill creation or edit, including small changes.
Its output must cover the target, expected behavior, fixtures, acceptance
criteria, baseline, feedback revisions, and validation. Use the references as
design evidence during planning; do not write target skill files then.
Planning alone does not authorize implementation. Obtain the required decisions
and implementation authorization, then invoke implement-plan with that plan.
An explicit request to implement an already approved plan supplies that
authorization within its scope.
When executing an already approved authoring task, verify that it covers this
change and consume these references inside that task. Do not recursively
restart the top-level skill or ask for an identical plan. Every logical change
still needs an approved plan and execution through implement-plan; there is
no small-edit exemption. New requirements or feedback outside the covered
revision return to planning before another edit.
An evaluation-only request that does not change a skill can use the evaluation reference directly within the request's authority.
Execute, evaluate, and review
Use keep-it-simple before authoring and cleanup. Write only the resources
needed for the agreed behavior. For improvement, preserve an immutable snapshot
before the first target edit. Agree on realistic prompts, expected outputs, and
verifiable expectations before running them.
Execute real target tasks and comparable independent baselines. Respect host capacity; serial fresh sessions are valid when child concurrency is unavailable. Grade actual outputs with grader guidance, then aggregate and inspect analysis guidance. Missing runs or evidence are incomplete, not passes. Never substitute characters for tokens or fabricate unavailable telemetry.
Generate the bundled static review artifact and present it before revising from feedback. Distinguish synthetic fixture feedback from human acceptance. Apply only revisions covered by the approved plan, rerun affected evaluations, and preserve previous iteration outputs. Stop at the plan's iteration limit; unsettled scope returns to planning.
Invoke review-pro-max on the authored change through implement-plan before
delivery. Follow its existing review/fix limits. Publication or installation
outside the approved scope requires separate authority; neither is implied by
authoring or evaluation.
Completion evidence
Return the skill location and identity, approved plan, produced artifacts, actual test prompts and outputs, baseline identity, grades and benchmark, human feedback status, review verdict, and remaining limitations. A plan, installed indicator, or helper exit status alone does not prove the target workflow. Report unsupported capabilities honestly; optional optimization must not block portable authoring or evaluation.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| 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: 467k | Updated | Claude CodeCodex | Verified | Install |
| 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: 319k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| 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 available | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |