Skill Creator

Skill Creator

A skill for creating new skills and iteratively improving them.

ENVIRONMENT: ScienceClaw — You are running in ScienceClaw. There are NO subagents, NO task tool, NO claude CLI. Do NOT attempt to spawn subagents or call task. For eval/testing, saving, and improving skills, follow the ScienceClaw-Specific Instructions section at the bottom of this file. The "Creating a skill" section below is universal and can be followed as-is. All other sections (Running and evaluating test cases, Improving the skill, Blind comparison, Description Optimization) are Claude Code-specific — SKIP them entirely and use the ScienceClaw equivalents instead.

Your job is to figure out where the user is in the skill creation process and help them progress. The high-level flow is: draft → test → review → improve → repeat.

• If the user wants to create a new skill, help narrow down what they mean, write a draft, then offer to test it
• If they already have a draft, go straight to the eval/iterate part
• If they just want a quick change, skip the full eval cycle
• Always be flexible — if the user says "just vibe with me", do that instead

Communicating with the user

The skill creator is liable to be used by people across a wide range of familiarity with coding jargon. If you haven't heard (and how could you, it's only very recently that it started), there's a trend now where the power of Claude is inspiring plumbers to open up their terminals, parents and grandparents to google "how to install npm". On the other hand, the bulk of users are probably fairly computer-literate.

So please pay attention to context cues to understand how to phrase your communication! In the default case, just to give you some idea:

• "evaluation" and "benchmark" are borderline, but OK
• for "JSON" and "assertion" you want to see serious cues from the user that they know what those things are before using them without explaining them

It's OK to briefly explain terms if you're in doubt, and feel free to clarify terms with a short definition if you're unsure if the user will get it.

Creating a skill

Capture Intent

Start by understanding the user's intent. The current conversation might already contain a workflow the user wants to capture (e.g., they say "turn this into a skill"). If so, extract answers from the conversation history first — the tools used, the sequence of steps, corrections the user made, input/output formats observed. The user may need to fill the gaps, and should confirm before proceeding to the next step.

1. What should this skill enable Claude to do?
2. When should this skill trigger? (what user phrases/contexts)
3. What's the expected output format?
4. Should we set up test cases to verify the skill works? Skills with objectively verifiable outputs (file transforms, data extraction, code generation, fixed workflow steps) benefit from test cases. Skills with subjective outputs (writing style, art) often don't need them. Suggest the appropriate default based on the skill type, but let the user decide.

Interview and Research

Proactively ask questions about edge cases, input/output formats, example files, success criteria, and dependencies. Wait to write test prompts until you've got this part ironed out.

Use web_search and web_crawl to research. When the skill involves external APIs, services, data sources, or technical workflows you're not 100% sure about, use seekr_sdk (from seekr_sdk import web_search, web_crawl) to verify current information before writing instructions. This includes:

• Confirming API endpoints, parameters, and response formats are still valid
• Looking up best practices, official documentation, or community patterns
• Finding similar skills or workflows for inspiration
• Verifying that libraries, tools, or services mentioned in the skill actually exist and work as described

Come prepared with context to reduce burden on the user.

Write the SKILL.md

YAML Frontmatter is MANDATORY. Every SKILL.md MUST start with YAML frontmatter wrapped in --- fences containing name and description. Without frontmatter, the skill will have an empty description in the "Available Skills" list, and the agent will never know what the skill does — it simply won't be triggered. This is the #1 cause of skills not working after creation.

The SKILL.md must follow this exact structure:

---
name: my-skill-name
description: "What this skill does and when to trigger it. Be specific and pushy."
---

# my-skill-name

(rest of the skill body — instructions, examples, etc.)

Frontmatter fields:

• name (required): Skill identifier, should match the directory name.
• description (required): This is the primary triggering mechanism — the agent reads this to decide whether to use the skill. Include both what the skill does AND specific contexts/phrases for when to use it. All "when to use" info goes here, not in the body. Make the description "pushy" — err on the side of triggering too often rather than too rarely. For instance, instead of "Build a dashboard.", write "Build dashboards for data visualization. Use this whenever the user mentions dashboards, charts, metrics display, data overview, reporting, or wants to visualize any kind of data, even if they don't explicitly ask for a 'dashboard.'" Include both Chinese and English trigger phrases if the skill may be used in bilingual contexts.
• compatibility (optional): Required tools, dependencies (rarely needed).

Then write the rest of the skill body (instructions, examples, output format, etc.).

Skill Writing Guide

Anatomy of a Skill

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter (--- name + description ---, MANDATORY or skill won't trigger)
│   └── Markdown body (instructions, examples, output format)
└── Bundled Resources (optional)
    ├── scripts/    - Executable code for deterministic/repetitive tasks
    ├── references/ - Docs loaded into context as needed
    └── assets/     - Files used in output (templates, icons, fonts)

Progressive Disclosure

Skills use a three-level loading system:

1. Metadata (name + description) - Always in context (~100 words)
2. SKILL.md body - In context whenever skill triggers (<500 lines ideal)
3. Bundled resources - As needed (unlimited, scripts can execute without loading)

These word counts are approximate and you can feel free to go longer if needed.

Key patterns:

• Keep SKILL.md under 500 lines; if you're approaching this limit, add an additional layer of hierarchy along with clear pointers about where the model using the skill should go next to follow up.
• Reference files clearly from SKILL.md with guidance on when to read them
• For large reference files (>300 lines), include a table of contents

Domain organization: When a skill supports multiple domains/frameworks, organize by variant:

cloud-deploy/
├── SKILL.md (workflow + selection)
└── references/
    ├── aws.md
    ├── gcp.md
    └── azure.md

Claude reads only the relevant reference file.

Principle of Lack of Surprise

This goes without saying, but skills must not contain malware, exploit code, or any content that could compromise system security. A skill's contents should not surprise the user in their intent if described. Don't go along with requests to create misleading skills or skills designed to facilitate unauthorized access, data exfiltration, or other malicious activities. Things like a "roleplay as an XYZ" are OK though.

Writing Patterns

Prefer using the imperative form in instructions.

Defining output formats - You can do it like this:

## Report structure
ALWAYS use this exact template:
# [Title]
## Executive summary
## Key findings
## Recommendations

Examples pattern - It's useful to include examples. You can format them like this (but if "Input" and "Output" are in the examples you might want to deviate a little):

## Commit message format
**Example 1:**
Input: Added user authentication with JWT tokens
Output: feat(auth): implement JWT-based authentication

Writing Style

Try to explain to the model why things are important in lieu of heavy-handed musty MUSTs. Use theory of mind and try to make the skill general and not super-narrow to specific examples. Start by writing a draft and then look at it with fresh eyes and improve it.

Test Cases

After writing the skill draft, come up with 2-3 realistic test prompts — the kind of thing a real user would actually say. Share them with the user: [you don't have to use this exact language] "Here are a few test cases I'd like to try. Do these look right, or do you want to add more?" Then run them.

Save test cases to evals/evals.json. Don't write assertions yet — just the prompts. You'll draft assertions in the next step while the runs are in progress.

{
  "skill_name": "example-skill",
  "evals": [
    {
      "id": 1,
      "prompt": "User's task prompt",
      "expected_output": "Description of expected result",
      "files": []
    }
  ]
}

See references/schemas.md for the full schema (including the assertions field, which you'll add later).

Running and evaluating test cases

⚠️ SCIENCECLAW: SKIP THIS ENTIRE SECTION. This section requires subagents and task tool which are NOT available. Go to ScienceClaw-Specific Instructions → Eval / Improve Cycle instead.

This section is one continuous sequence — don't stop partway through. Do NOT use /skill-test or any other testing skill.

Put results in <skill-name>-workspace/ as a sibling to the skill directory. Within the workspace, organize results by iteration (iteration-1/, iteration-2/, etc.) and within that, each test case gets a directory (eval-0/, eval-1/, etc.). Don't create all of this upfront — just create directories as you go.

Step 1: Spawn all runs (with-skill AND baseline) in the same turn

For each test case, spawn two subagents in the same turn — one with the skill, one without. This is important: don't spawn the with-skill runs first and then come back for baselines later. Launch everything at once so it all finishes around the same time.

With-skill run:

Execute this task:
- Skill path: <path-to-skill>
- Task: <eval prompt>
- Input files: <eval files if any, or "none">
- Save outputs to: <workspace>/iteration-<N>/eval-<ID>/with_skill/outputs/
- Outputs to save: <what the user cares about — e.g., "the .docx file", "the final CSV">

Baseline run (same prompt, but the baseline depends on context):

• Creating a new skill: no skill at all. Same prompt, no skill path, save to without_skill/outputs/.
• Improving an existing skill: the old version. Before editing, snapshot the skill (cp -r <skill-path> <workspace>/skill-snapshot/), then point the baseline subagent at the snapshot. Save to old_skill/outputs/.

Write an eval_metadata.json for each test case (assertions can be empty for now). Give each eval a descriptive name based on what it's testing — not just "eval-0". Use this name for the directory too. If this iteration uses new or modified eval prompts, create these files for each new eval directory — don't assume they carry over from previous iterations.

{
  "eval_id": 0,
  "eval_name": "descriptive-name-here",
  "prompt": "The user's task prompt",
  "assertions": []
}

Step 2: While runs are in progress, draft assertions

Don't just wait for the runs to finish — you can use this time productively. Draft quantitative assertions for each test case and explain them to the user. If assertions already exist in evals/evals.json, review them and explain what they check.

Good assertions are objectively verifiable and have descriptive names — they should read clearly in the benchmark viewer so someone glancing at the results immediately understands what each one checks. Subjective skills (writing style, design quality) are better evaluated qualitatively — don't force assertions onto things that need human judgment.

Update the eval_metadata.json files and evals/evals.json with the assertions once drafted. Also explain to the user what they'll see in the viewer — both the qualitative outputs and the quantitative benchmark.

Step 3: As runs complete, capture timing data

When each subagent task completes, you receive a notification containing total_tokens and duration_ms. Save this data immediately to timing.json in the run directory:

{
  "total_tokens": 84852,
  "duration_ms": 23332,
  "total_duration_seconds": 23.3
}

This is the only opportunity to capture this data — it comes through the task notification and isn't persisted elsewhere. Process each notification as it arrives rather than trying to batch them.

Step 4: Grade, aggregate, and launch the viewer

Once all runs are done:

1. Grade each run — spawn a grader subagent (or grade inline) that reads agents/grader.md and evaluates each assertion against the outputs. Save results to grading.json in each run directory. The grading.json expectations array must use the fields text, passed, and evidence (not name/met/details or other variants) — the viewer depends on these exact field names. For assertions that can be checked programmatically, write and run a script rather than eyeballing it — scripts are faster, more reliable, and can be reused across iterations.

2. Aggregate into benchmark — run the aggregation script from the skill-creator directory:

   python -m scripts.aggregate_benchmark <workspace>/iteration-N --skill-name <name>
   

   This produces benchmark.json and benchmark.md with pass_rate, time, and tokens for each configuration, with mean ± stddev and the delta. If generating benchmark.json manually, see references/schemas.md for the exact schema the viewer expects. Put each with_skill version before its baseline counterpart.

3. Do an analyst pass — read the benchmark data and surface patterns the aggregate stats might hide. See agents/analyzer.md (the "Analyzing Benchmark Results" section) for what to look for — things like assertions that always pass regardless of skill (non-discriminating), high-variance evals (possibly flaky), and time/token tradeoffs.

4. Launch the viewer with both qualitative outputs and quantitative data:

   nohup python <skill-creator-path>/eval-viewer/generate_review.py \
     <workspace>/iteration-N \
     --skill-name "my-skill" \
     --benchmark <workspace>/iteration-N/benchmark.json \
     > /dev/null 2>&1 &
   VIEWER_PID=$!
   

   For iteration 2+, also pass --previous-workspace <workspace>/iteration-<N-1>.

   Cowork / headless environments: If webbrowser.open() is not available or the environment has no display, use --static <output_path> to write a standalone HTML file instead of starting a server. Feedback will be downloaded as a feedback.json file when the user clicks "Submit All Reviews". After download, copy feedback.json into the workspace directory for the next iteration to pick up.

Note: please use generate_review.py to create the viewer; there's no need to write custom HTML.

5. Tell the user something like: "I've opened the results in your browser. There are two tabs — 'Outputs' lets you click through each test case and leave feedback, 'Benchmark' shows the quantitative comparison. When you're done, come back here and let me know."

What the user sees in the viewer

The "Outputs" tab shows one test case at a time:

• Prompt: the task that was given
• Output: the files the skill produced, rendered inline where possible
• Previous Output (iteration 2+): collapsed section showing last iteration's output
• Formal Grades (if grading was run): collapsed section showing assertion pass/fail
• Feedback: a textbox that auto-saves as they type
• Previous Feedback (iteration 2+): their comments from last time, shown below the textbox

The "Benchmark" tab shows the stats summary: pass rates, timing, and token usage for each configuration, with per-eval breakdowns and analyst observations.

Navigation is via prev/next buttons or arrow keys. When done, they click "Submit All Reviews" which saves all feedback to feedback.json.

Step 5: Read the feedback

When the user tells you they're done, read feedback.json:

{
  "reviews": [
    {"run_id": "eval-0-with_skill", "feedback": "the chart is missing axis labels", "timestamp": "..."},
    {"run_id": "eval-1-with_skill", "feedback": "", "timestamp": "..."},
    {"run_id": "eval-2-with_skill", "feedback": "perfect, love this", "timestamp": "..."}
  ],
  "status": "complete"
}

Empty feedback means the user thought it was fine. Focus your improvements on the test cases where the user had specific complaints.

Kill the viewer server when you're done with it:

kill $VIEWER_PID 2>/dev/null

Improving the skill

⚠️ SCIENCECLAW: SKIP THIS SECTION. Go to ScienceClaw-Specific Instructions → Eval / Improve Cycle → Step 5 instead.

This is the heart of the loop. You've run the test cases, the user has reviewed the results, and now you need to make the skill better based on their feedback.

How to think about improvements

1. Generalize from the feedback. The big picture thing that's happening here is that we're trying to create skills that can be used a million times (maybe literally, maybe even more who knows) across many different prompts. Here you and the user are iterating on only a few examples over and over again because it helps move faster. The user knows these examples in and out and it's quick for them to assess new outputs. But if the skill you and the user are codeveloping works only for those examples, it's useless. Rather than put in fiddly overfitty changes, or oppressively constrictive MUSTs, if there's some stubborn issue, you might try branching out and using different metaphors, or recommending different patterns of working. It's relatively cheap to try and maybe you'll land on something great.

2. Keep the prompt lean. Remove things that aren't pulling their weight. Make sure to read the transcripts, not just the final outputs — if it looks like the skill is making the model waste a bunch of time doing things that are unproductive, you can try getting rid of the parts of the skill that are making it do that and seeing what happens.

3. Explain the why. Try hard to explain the why behind everything you're asking the model to do. Today's LLMs are smart. They have good theory of mind and when given a good harness can go beyond rote instructions and really make things happen. Even if the feedback from the user is terse or frustrated, try to actually understand the task and why the user is writing what they wrote, and what they actually wrote, and then transmit this understanding into the instructions. If you find yourself writing ALWAYS or NEVER in all caps, or using super rigid structures, that's a yellow flag — if possible, reframe and explain the reasoning so that the model understands why the thing you're asking for is important. That's a more humane, powerful, and effective approach.

4. Look for repeated work across test cases. Read the transcripts from the test runs and notice if the subagents all independently wrote similar helper scripts or took the same multi-step approach to something. If all 3 test cases resulted in the subagent writing a create_docx.py or a build_chart.py, that's a strong signal the skill should bundle that script. Write it once, put it in scripts/, and tell the skill to use it. This saves every future invocation from reinventing the wheel.

This task is pretty important (we are trying to create billions a year in economic value here!) and your thinking time is not the blocker; take your time and really mull things over. I'd suggest writing a draft revision and then looking at it anew and making improvements. Really do your best to get into the head of the user and understand what they want and need.

The iteration loop

After improving the skill:

1. Apply your improvements to the skill
2. Rerun all test cases into a new iteration-<N+1>/ directory, including baseline runs. If you're creating a new skill, the baseline is always without_skill (no skill) — that stays the same across iterations. If you're improving an existing skill, use your judgment on what makes sense as the baseline: the original version the user came in with, or the previous iteration.
3. Launch the reviewer with --previous-workspace pointing at the previous iteration
4. Wait for the user to review and tell you they're done
5. Read the new feedback, improve again, repeat

Keep going until:

• The user says they're happy
• The feedback is all empty (everything looks good)
• You're not making meaningful progress

Advanced: Blind comparison

⚠️ SCIENCECLAW: SKIP. Blind comparison requires subagents which are not available.

For situations where you want a more rigorous comparison between two versions of a skill (e.g., the user asks "is the new version actually better?"), there's a blind comparison system. Read agents/comparator.md and agents/analyzer.md for the details. The basic idea is: give two outputs to an independent agent without telling it which is which, and let it judge quality. Then analyze why the winner won.

This is optional, requires subagents, and most users won't need it. The human review loop is usually sufficient.

Description Optimization

⚠️ SCIENCECLAW: SKIP THIS SECTION. The automated run_loop.py / claude -p approach is not available. Go to ScienceClaw-Specific Instructions → Description Optimization for manual optimization guidance.

The description field in SKILL.md frontmatter is the primary mechanism that determines whether Claude invokes a skill. After creating or improving a skill, offer to optimize the description for better triggering accuracy.

Step 1: Generate trigger eval queries

Create 20 eval queries — a mix of should-trigger and should-not-trigger. Save as JSON:

[
  {"query": "the user prompt", "should_trigger": true},
  {"query": "another prompt", "should_trigger": false}
]

The queries must be realistic and something a Claude Code or Claude.ai user would actually type. Not abstract requests, but requests that are concrete and specific and have a good amount of detail. For instance, file paths, personal context about the user's job or situation, column names and values, company names, URLs. A little bit of backstory. Some might be in lowercase or contain abbreviations or typos or casual speech. Use a mix of different lengths, and focus on edge cases rather than making them clear-cut (the user will get a chance to sign off on them).

Bad: "Format this data", "Extract text from PDF", "Create a chart"

Good: "ok so my boss just sent me this xlsx file (its in my downloads, called something like 'Q4 sales final FINAL v2.xlsx') and she wants me to add a column that shows the profit margin as a percentage. The revenue is in column C and costs are in column D i think"

For the should-trigger queries (8-10), think about coverage. You want different phrasings of the same intent — some formal, some casual. Include cases where the user doesn't explicitly name the skill or file type but clearly needs it. Throw in some uncommon use cases and cases where this skill competes with another but should win.

For the should-not-trigger queries (8-10), the most valuable ones are the near-misses — queries that share keywords or concepts with the skill but actually need something different. Think adjacent domains, ambiguous phrasing where a naive keyword match would trigger but shouldn't, and cases where the query touches on something the skill does but in a context where another tool is more appropriate.

The key thing to avoid: don't make should-not-trigger queries obviously irrelevant. "Write a fibonacci function" as a negative test for a PDF skill is too easy — it doesn't test anything. The negative cases should be genuinely tricky.

Step 2: Review with user

Present the eval set to the user for review using the HTML template:

1. Read the template from assets/eval_review.html
2. Replace the placeholders:
   • __EVAL_DATA_PLACEHOLDER__ → the JSON array of eval items (no quotes around it — it's a JS variable assignment)
   • __SKILL_NAME_PLACEHOLDER__ → the skill's name
   • __SKILL_DESCRIPTION_PLACEHOLDER__ → the skill's current description
3. Write to a temp file (e.g., /tmp/eval_review_<skill-name>.html) and open it: open /tmp/eval_review_<skill-name>.html
4. The user can edit queries, toggle should-trigger, add/remove entries, then click "Export Eval Set"
5. The file downloads to ~/Downloads/eval_set.json — check the Downloads folder for the most recent version in case there are multiple (e.g., eval_set (1).json)

This step matters — bad eval queries lead to bad descriptions.

Step 3: Run the optimization loop

Tell the user: "This will take some time — I'll run the optimization loop in the background and check on it periodically."

Save the eval set to the workspace, then run in the background:

python -m scripts.run_loop \
  --eval-set <path-to-trigger-eval.json> \
  --skill-path <path-to-skill> \
  --model <model-id-powering-this-session> \
  --max-iterations 5 \
  --verbose

Use the model ID from your system prompt (the one powering the current session) so the triggering test matches what the user actually experiences.

While it runs, periodically tail the output to give the user updates on which iteration it's on and what the scores look like.

This handles the full optimization loop automatically. It splits the eval set into 60% train and 40% held-out test, evaluates the current description (running each query 3 times to get a reliable trigger rate), then calls Claude with extended thinking to propose improvements based on what failed. It re-evaluates each new description on both train and test, iterating up to 5 times. When it's done, it opens an HTML report in the browser showing the results per iteration and returns JSON with best_description — selected by test score rather than train score to avoid overfitting.

How skill triggering works

Understanding the triggering mechanism helps design better eval queries. Skills appear in Claude's available_skills list with their name + description, and Claude decides whether to consult a skill based on that description. The important thing to know is that Claude only consults skills for tasks it can't easily handle on its own — simple, one-step queries like "read this PDF" may not trigger a skill even if the description matches perfectly, because Claude can handle them directly with basic tools. Complex, multi-step, or specialized queries reliably trigger skills when the description matches.

This means your eval queries should be substantive enough that Claude would actually benefit from consulting a skill. Simple queries like "read file X" are poor test cases — they won't trigger skills regardless of description quality.

Step 4: Apply the result

Take best_description from the JSON output and update the skill's SKILL.md frontmatter. Show the user before/after and report the scores.

Package and Present (only if present_files tool is available)

⚠️ SCIENCECLAW: SKIP. present_files and scripts.package_skill are not available.

Check whether you have access to the present_files tool. If you don't, skip this step. If you do, package the skill and present the .skill file to the user:

python -m scripts.package_skill <path/to/skill-folder>

After packaging, direct the user to the resulting .skill file path so they can install it.

ScienceClaw-Specific Instructions

You are running inside the ScienceClaw environment. The core skill-creation workflow (draft → test → review → improve → repeat) still applies, but with these environment-specific adaptations:

Environment Constraints

• No subagents: You cannot spawn parallel subagents. Run all test cases sequentially, one at a time.
• No claude CLI: The claude -p command is not available. Skip the automated description optimization loop (run_loop.py, run_eval.py, improve_description.py). Instead, iterate on the description manually with the user.
• No browser display: The sandbox has no display. Do NOT try to open HTML files. Instead, save generated HTML/results to the workspace directory — the user can view files through the frontend file viewer.
• Sandbox execution: All commands run in a remote sandbox container. Use absolute paths under the workspace directory.

Web Research Best Practices (CRITICAL)

When writing or improving a skill, the instructions you write may reference external APIs, services, libraries, or technical workflows. Do NOT assume URLs, endpoints, or API formats from memory — they may be outdated or wrong. Always verify via seekr_sdk first.

DO NOT blindly guess URLs or API endpoints. If you need to include API calls, code examples, or service references in a skill's instructions:

1. Search first: web_search("service_name API documentation 2026") to find current docs
2. Crawl the docs: web_crawl("<documentation_url>") to confirm endpoints, parameters, and response formats
3. Then write the instructions based on verified, working information

Anti-pattern (NEVER do this):

# Writing skill instructions with guessed URLs:
"Call https://api.example.com/v1/data to get ..."   # might be 404
"Use https://api.example.com/v2/endpoint ..."        # might not exist
# ... multiple guessed URL variations in the skill body ...

Correct pattern:

from seekr_sdk import web_search, web_crawl
# Verify the API/service before writing instructions
results = web_search("example.com API documentation current endpoint")
doc = web_crawl("https://docs.example.com/api-reference")
# Now write skill instructions with confirmed information

This applies to all phases — initial drafting, upgrading existing skills, and writing code examples within skills. External APIs and services change frequently; search engines exist for a reason.

Skill Save Workflow (CRITICAL)

After the skill is created (or modified) and the user is satisfied:

1. Write the skill files to the session workspace:

   • The workspace directory is provided in the system prompt (e.g., /home/scienceclaw/<session-id>/)
   • Create the directory: <workspace>/skills/<skill-name>/ (or <workspace>/.agents/skills/<skill-name>/)
   • The SKILL.md MUST include YAML frontmatter with name and description fields:

     ---
     name: my-skill
     description: "When to trigger and what this skill does..."
     ---
     
     # my-skill
     (rest of the skill body...)
     

   • Write any bundled resources (scripts/, references/, assets/) as needed

2. Call propose_skill_save: After writing all files, ALWAYS call the propose_skill_save tool:

   propose_skill_save(skill_name="<skill-name>")
   

   This triggers a UI prompt asking the user to save the skill permanently to their skill library.

3. NEVER write directly to /skills/ — that directory is read-only in the sandbox. Always go through the workspace → propose_skill_save flow.

Modifying an Existing Skill

When the user asks to modify or improve an existing skill:

1. Read the current skill from /skills/<skill-name>/SKILL.md
2. Copy / rewrite the modified version to the session workspace under <workspace>/skills/<skill-name>/
3. Follow the same save workflow above (propose_skill_save → user confirms → system copies to permanent location)

Eval / Improve Cycle

ScienceClaw provides two built-in tools — eval_skill and grade_eval — that replace manual sequential testing with independent agent sessions. Each test runs in a fresh agent that knows nothing about the test design, simulating a real user. The core loop is: draft → eval → grade → review → improve → re-eval.

Step 1: Design Test Cases

After drafting or modifying the skill, create 2-3 realistic test prompts. Confirm them with the user before running.

[
  {"id": "t1", "prompt": "Realistic user prompt here", "description": "Short label for this test"},
  {"id": "t2", "prompt": "Another realistic prompt", "description": "What this tests"}
]

Step 2: Run Eval

Call the eval_skill tool. It creates a temporary independent agent for each test case — this agent loads the skill and executes the prompt in a clean context (no knowledge of test design, no meta-tools).

eval_skill(
  workspace_dir="<your workspace dir>",
  skill_name="my-skill",
  test_cases_json='[{"id":"t1","prompt":"...","description":"..."},...]',
  iteration=1
)

The tool runs all tests sequentially and saves outputs to <workspace>/<skill-name>-eval/iteration-<N>/. Each test case gets:

• prompt.txt — the input
• response.md — the agent's final response
• tool_calls.json — all tool calls made
• meta.json — timing and token data

It returns a text summary with pass/fail status, duration, and response previews.

For iteration 2+, it automatically loads the previous iteration's results and includes a comparison in the summary (status changes, duration deltas).

Step 3: Grade Results

After eval completes, use grade_eval to run programmatic assertions:

grade_eval(
  eval_dir="<workspace>/<skill-name>-eval/iteration-1",
  assertions_json='[
    {"test_id":"t1", "type":"response_contains", "value":"hello", "description":"Should greet user"},
    {"test_id":"t1", "type":"min_tool_calls", "value":"2", "description":"Should use at least 2 tools"},
    {"test_id":"t2", "type":"json_valid", "value":"", "description":"Output should be valid JSON"}
  ]'
)

Supported assertion types:

• file_exists — a file was created in the eval output
• response_contains / response_not_contains — substring check
• json_valid — valid JSON (optionally check a specific file)
• regex_match — regex pattern match
• min_tool_calls — minimum number of tool calls
• tool_was_used — a specific tool was called

For subjective assertions (writing quality, design aesthetics), read the response.md files yourself and assess — don't force them into programmatic checks.

The tool writes grading.json and returns a pass/fail summary.

Step 4: Present Results to User

Show the user a concise summary:

• Per test: prompt, pass/fail, key observations, where output files are
• Overall: X/Y assertions passed, notable patterns

Then ask: "How do these look? Anything you'd change?"

Point users to <workspace>/<skill-name>-eval/iteration-<N>/ to browse detailed outputs via the file viewer.

Step 5: Improve the Skill

Based on grading results and user feedback:

1. Generalize — don't overfit to the test cases. Fix issues in a way that works for all inputs.
2. Trim — remove instructions that didn't help or caused wasted steps.
3. Explain why — convey reasoning rather than adding rigid rules.
4. Bundle repeated work — if every test run wrote the same helper script, put it in scripts/.

Apply improvements, then re-eval with iteration=<N+1>. The comparison with the previous iteration is automatic. Keep going until:

• The user is happy
• All assertions pass
• You're not making meaningful progress

Quick Iteration (for simple changes)

If the user just wants a small tweak (fix a typo, adjust output format), skip the full eval cycle:

1. Read → modify → write to workspace
2. Eyeball the change
3. propose_skill_save

Reserve the full eval cycle for substantive changes where correctness matters.

Retroactive Skill Creation (from a completed task)

Sometimes the agent completes a multi-step task and realizes (or the user requests) that the workflow should be captured as a reusable skill. This is the "reflect & capture" mode.

When to trigger:

• The agent just completed a non-trivial task (5+ tool calls, multi-step workflow)
• The user says "turn this into a skill", "save this workflow", or similar
• The system prompt's Step 4 (Reflect) suggests capturing the workflow

How to create a skill from a completed task:

1. Review the conversation history: Look at the sequence of tool calls, decisions, and outputs. Identify the core workflow steps.
2. Generalize: Replace session-specific values (file paths, query strings, URLs) with parameters or placeholders. Extract the reusable pattern.
3. Write the SKILL.md: Structure the workflow as clear instructions with:
   • YAML frontmatter (name + description)
   • Step-by-step workflow (the generalized version of what you just did)
   • Decision points (where you made choices, explain the criteria)
   • Expected inputs/outputs for each step
4. Save: Write to workspace → propose_skill_save

Example: If you just completed a "literature review" task by searching → crawling 5 papers → extracting key findings → writing a summary, the skill would capture this as a reusable "literature-review" workflow with parameterized topic, search queries, and output format.

Description Optimization

The description field in YAML frontmatter is the primary trigger mechanism — it determines whether the agent reads the skill. Optimize it manually:

1. Draft a description that covers what the skill does AND when to use it
2. Make it "pushy" — err on the side of triggering too often rather than too rarely. Example:
   • Weak: "Create a dashboard"
   • Strong: "Create dashboards for data visualization. Use this whenever the user mentions dashboards, charts, metrics display, data overview, reporting, or wants to visualize any kind of data."
3. Walk through 3-4 realistic prompts mentally: would this description make the agent read the skill? Adjust until it reliably triggers.

Common Pitfalls

1. Missing YAML frontmatter: This is the #1 cause of skills not working. Every SKILL.md must start with --- fenced YAML containing name and description. Without it, the skill appears in the Available Skills list with an empty description, and the agent will never know when to use it — effectively making the skill invisible. Always verify the frontmatter is present before saving.
2. Vague or missing description: Even with frontmatter, a weak description like "A useful skill" won't trigger reliably. The description must clearly state what the skill does AND list specific trigger phrases/contexts. Include both Chinese and English terms if the skill may be used bilingually.
3. Writing directly to /skills/: The skills directory is read-only in the sandbox. Always write to the session workspace first, then use propose_skill_save.
4. Guessing URLs or API details in instructions: When skill instructions reference external APIs, services, or documentation, verify them via web_search / web_crawl first. Don't write instructions with assumed URLs — they may be outdated or wrong.
5. Overly rigid instructions: Prefer explaining why over heavy-handed MUST/NEVER rules. The agent using the skill is intelligent — convey reasoning so it can adapt to edge cases.
6. SKILL.md too long: Keep under 500 lines. If approaching this limit, split into references/ files with clear pointers from the main SKILL.md.