AgentSkillsCN

experience-crystallizer

通过分析当前会话的历史记录，从失败中汲取经验，提炼并固化新的技能。当您历经反复尝试与摸索，成功解决了一个棘手问题，并希望将这一解决方案保存下来，以备日后参考时，可使用此技能。

SKILL.md
--- frontmatter
name: experience-crystallizer
description: Analyze current session history to learn from mistakes and crystallize a new skill. Use this when you successfully solved a hard problem after trial and error, and want to save the solution for future use.
tools:
  - name: find_current_session_context
    description: Locate the physical path of the current session's log file (context.jsonl). MUST be called with a unique fingerprint string.
    parameters:
      properties:
        fingerprint:
          description: A unique random string (e.g. UUID) that you MUST generate and include in this tool call.
          type: string
      required:
        - fingerprint
    command: /Users/moonshot/.local/share/uv/tools/kimi-cli/bin/python3 skills/experience-crystallizer/scripts/locate_context.py {fingerprint}

  - name: analyze_session_log
    description: Read and summarize the session log to extract user intent, errors, and successful actions.
    parameters:
      properties:
        path:
          description: The absolute path to the context.jsonl file.
          type: string
      required:
        - path
    command: /Users/moonshot/.local/share/uv/tools/kimi-cli/bin/python3 skills/experience-crystallizer/scripts/analyze_log.py {path}

  - name: validate_generated_skill
    description: Check if the generated skill complies with the Agent Skills specification.
    parameters:
      properties:
        path:
          description: The absolute path to the directory of the newly created skill.
          type: string
      required:
        - path
    command: /Users/moonshot/.local/share/uv/tools/kimi-cli/bin/python3 skills/experience-crystallizer/scripts/validate_skill.py {path}

Experience Crystallizer

This skill enables you to perform Self-Learning by creating standardized skills compliant with the Agent Skills Specification.

Trigger Criteria (WHEN TO USE)

CRITICAL: Do NOT use this skill for trivial tasks.

You should ONLY use this skill if the current session meets at least one of these criteria:

  1. •Trial & Error: You failed at least once (e.g., SyntaxError, logic error, wrong API usage) and had to fix it.
  2. •Discovery: You had to read external documentation or search the web to find a non-obvious solution.
  3. •Complexity: The solution required multiple steps or specific configurations that are hard to remember.

If the task was completed successfully on the first try with common knowledge, DO NOT create a skill.

Workflow (Decision Flow)

mermaid
graph TD
    A[Start] --> B[Generate Fingerprint]
    B --> C[Locate Context]
    C --> D[Analyze Log]
    D --> E[Identify Solution]
    E --> F[Generate Skill Files]
    F --> G[Validate Skill]
    G -->|Pass| H[Done]
    G -->|Fail| I[Fix & Retry]
    I --> G

Output Specification (Agent Skills Standard)

You must create the new skill in skills/learned/<skill-name>/ with the following structure.

1. Directory Structure

text
skills/learned/<skill-name>/
  ├── SKILL.md            # [Required] Entry point
  ├── scripts/            # [Optional] Executable tools (was helpers/)
  ├── references/         # [Optional] Detailed docs
  └── assets/             # [Optional] Static resources
      └── examples/       # [Recommended] Read-only code (.example)

2. Validation Rules (Strict)

  • •Name: SKILL.md frontmatter name must match the directory name. Lowercase, hyphens only.
  • •Code Isolation: Files in assets/examples/ MUST end with .example.
  • •Executable: Place agent tools in scripts/.
  • •Documentation:
    • •Keep SKILL.md concise (< 500 lines).
    • •Use Mermaid flowcharts for logic.
    • •Include ## Common Pitfalls.

Step-by-Step Implementation

  1. •Read log analysis.
  2. •Determine Root Cause & Solution.
  3. •Write SKILL.md and auxiliary files.
  4. •RUN validate_generated_skill. Fix any errors reported.