nguyentien06ck3/agent-skill-innovation
Overview
This skill detects improvement opportunities after a task completes, encounters blockers, or receives quality feedback and produces a concise Innovation Retrospective. It summarizes observed signals, hypothesizes root causes, and recommends whether to update an existing skill or create a new one. It always asks for explicit confirmation before invoking the skill-creation workflow.
How this skill works
The skill monitors end-of-task conditions, error states, and explicit user feedback to collect signals such as negative sentiment, repeated clarifications, tool failures, or suggested workflow changes. It generates a short retrospective with observed signals, 1–3 root-cause hypotheses, and a clear recommendation: update an existing skill or create a new one. If the user confirms, it runs a defined skill-creator workflow to prepare a change package, but never executes that workflow without explicit permission.
When to use it
- After delivering a final output or when the conversation clearly ends
- When tasks are blocked by errors, tool failures, or unresolved uncertainty
- When the user requests rework, expresses dissatisfaction, or notes missing requirements
- When a task repeats often or the same friction appears across conversations
- When a user suggests a better template, workflow, or delivery model
Best practices
- Keep retrospectives brief (8–20 lines) and evidence-focused; avoid long essays
- Paraphrase signals concisely and link them to specific examples from the exchange
- Prefer updating an existing skill when scope is a clarification or added guardrails
- Propose creating a new skill when the pattern recurs or requires a standardized flow
- Always request explicit user confirmation before preparing or executing any skill-creation steps
Example use cases
- User asks for a refactor after delivery; retrospective identifies missing edge cases and recommends updating the formatter skill
- Tool integration fails repeatedly; retrospective diagnoses tooling guardrails missing and recommends a new troubleshooting skill
- Multiple clarification loops occur; retrospective flags process friction and suggests adding planning prompts to an existing task skill
- User suggests a new output template used across teams; retrospective recommends creating a reusable template skill
- Final QA feedback cites incorrect facts; retrospective points to data validation gaps and proposes skill updates with verification steps
FAQ
No. It will never modify or create skills without your explicit confirmation to run the skill-creator workflow.
How long is the retrospective?
By default it is concise—typically 8–20 lines—covering signals, root-cause hypotheses, recommendation, and a proposed change summary.
How do you decide between updating versus creating a skill?
I recommend an update when the issue fits within an existing skill’s scope and needs clarifications or guardrails; I recommend creating a new skill when a recurring pattern or new standardized flow is needed.
