Evaluate agent definition files for purpose alignment and propose targeted edits.
npx skills add https://github.com/cosmicdreams/claude-plugins --skill self-cosmicdreams
Please help me install this Agent Skill. Skill: self Source: https://github.com/cosmicdreams/claude-plugins/tree/main/improve/skills/self Command: npx skills add https://github.com/cosmicdreams/claude-plugins --skill self-cosmicdreams
This Skill helps you detect when an agent’s prompt, model, tools, and instructions no longer match what it is supposed to achieve, so you can restore clear role alignment and effectiveness.
Use the self skill to evaluate an agent definition file and propose concrete revisions to fix any misalignment with its stated purpose.
High-intent search queries and answers about installing and using this skill.
To audit an agent definition, evaluate its frontmatter and prompt body to derive a core purpose, then check alignment across model fit, tool correctness, prompt clarity, scope, and actionability to propose targeted improvements.
Agent evaluation is the process of assessing whether an agent's model, tools, prompt, and instructions effectively match its intended goals, detecting drift or underperformance to restore clear role alignment and effectiveness.
Fix agent drift by deriving the agent's one-sentence purpose from its definition, identifying misalignments across key dimensions, and applying dimension-by-dimension proposed edits with rationale and trust-model application guidance.
Yes, an agent can perform a self-audit using the improve:self command to read its own full definition, evaluate its purpose alignment, and identify methodology gaps or instruction clarity issues.
An agent prompt audit checks alignment across model fit, tool correctness, prompt length and clarity, scope, actionability, and description quality to ensure the definition matches its stated purpose.
You should redesign an agent when evaluation reveals significant misalignment across multiple dimensions like model fit and tool correctness, requiring targeted improvements that go beyond simple prompt clarity adjustments.
cosmicdreamsAudit Claude Code agent markdown files and assign cost-efficient model tiers.
imbue-aiGenerate software specifications and design docs using a 13-step process.
vinitgirdharIntegrate and optimize the Anthropic Claude API across Python and TypeScript.
gabriellpequenoRun lint, type checks, and validations for Node.js/TypeScript and Python projects.
