--- name: skill-creator description: Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. --- # Skill Creator Create new skills and iteratively improve them. This copy runs natively on DeepSeek Harness. Skill triggering is measured by observing subagents in their real runtime and reading their session logs. At a high level, the process of creating a skill goes like this: - Decide what you want the skill to do and roughly how it should do it - Write a draft of the skill - Create a few test prompts and run subagents that have access to the skill on them - Help the user evaluate the results both qualitatively and quantitatively - While the runs happen in the background, draft some quantitative evals if there aren't any (if there are some, you can either use as is or modify if you feel something needs to change about them). Then explain them to the user (or if they already existed, explain the ones that already exist) - Use the `eval-viewer/generate_review.py` script to show the user the results for them to look at, and also let them look at the quantitative metrics - Rewrite the skill based on feedback from the user's evaluation of the results (and also if there are any glaring flaws that become apparent from the quantitative benchmarks) - Repeat until you're satisfied - Expand the test set and try again at larger scale Your job when using this skill is to figure out where the user is in this process and then jump in and help them progress through these stages. So for instance, maybe they're like "I want to make a skill for X". You can help narrow down what they mean, write a draft, write the test cases, figure out how they want to evaluate, run all the prompts, and repeat. On the other hand, maybe they already have a draft of the skill. In this case you can go straight to the eval/iterate part of the loop. Of course, you should always be flexible and if the user is like "I don't need to run a bunch of evaluations, just vibe with me", you can do that instead. Then after the skill is done (but again, the order is flexible), you can also run the description optimization loop to improve the skill's triggering. Cool? Cool. ## Communicating with the user The skill creator is liable to be used by people across a wide range of familiarity with coding jargon. The bulk of users are probably fairly computer-literate, and some are opening a terminal for the first time. 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 the agent 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. Check available MCPs - if useful for research (searching docs, finding similar skills, looking up best practices), research in parallel via subagents if available, otherwise inline. Come prepared with context to reduce burden on the user. ### Write the SKILL.md Based on the user interview, fill in these components: - **name**: Skill identifier - **description**: When to trigger, what it does. This is the primary triggering mechanism - include both what the skill does AND specific contexts for when to use it. All "when to use" info goes here, not in the body. Note: models have a tendency to "undertrigger" skills -- to not use them when they'd be useful. To combat this, please make the skill descriptions a little bit "pushy". So for instance, instead of "How to build a simple fast dashboard to display internal metrics.", you might write "How to build a simple fast dashboard to display internal metrics. Make sure to use this skill whenever the user mentions dashboards, data visualization, internal metrics, or wants to display any kind of company data, even if they don't explicitly ask for a 'dashboard.'" - **compatibility**: Required tools, dependencies (optional, rarely needed) - **the rest of the skill :)** ### Skill Writing Guide #### Anatomy of a Skill ``` skill-name/ ├── SKILL.md (required) │ ├── YAML frontmatter (name, description required) │ └── Markdown instructions └── 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 ``` The agent 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: ```markdown ## 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): ```markdown ## 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. ### Description quality review Before measuring anything, review the description against this checklist and fix what fails: - **Trigger phrases**: does it include the specific phrases users would actually say? - **Third person**: "This skill should be used when..." rather than "Load this skill when..." - **Specificity**: concrete scenarios, not vague categories - **Length**: not too short (under 50 characters), not too long (over 500) - **Anti-undertriggering**: the "pushy" wording from the frontmatter guidance — name the contexts that should trigger it even when the user doesn't name the skill ### 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. ```json { "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 This section is one continuous sequence — don't stop partway through. Do not hand this testing phase to another testing skill; the steps below are the process. Put results in `-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: - Task: - Input files: - Save outputs to: /iteration-/eval-/with_skill/outputs/ - Outputs to save: ``` **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-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. ```json { "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: Capture timing and tokens Per-subagent totals are not pushed to you. Do not assume they are lost, and do not try to catch them from a notification: in DSH they are persisted in each subagent's own session log, and can be read back at any time: ```bash python -m scripts.run_eval check --skill ``` The `seconds`, `total_tokens`, and `tool_calls` fields come from that run's log (`assistant/message` carries per-request `usage`). Do it after the runs settle. When you aggregate, remember that total context is re-sent every step, so summed `total_tokens` is dominated by cache reads — report output tokens alongside it rather than quoting a single number. ### 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 this skill's directory: ```bash python -m scripts.aggregate_benchmark /iteration-N --skill-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. The aggregation script only finds runs laid out as `eval-*//run-*/grading.json`. If you named the eval directories descriptively (which the guidance above tells you to do), it will report "No eval directories found" and a zero delta — that is a layout mismatch, not a measurement. Either add the `run-N/` level, or build `benchmark.json` yourself from the `grading.json` files following the schema exactly; the field names are load-bearing and a wrong one silently renders as zero. 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. Start it as a managed background job and give the user the URL; do not leave an unmanaged `nohup` process behind, and stop it when they are done: ```bash python /eval-viewer/generate_review.py \ /iteration-N \ --skill-name "my-skill" \ --benchmark /iteration-N/benchmark.json ``` For iteration 2+, also pass `--previous-workspace /iteration-`. **Headless / no display:** If `webbrowser.open()` is not available or the environment has no display, use `--static ` 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`: ```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. Stop the viewer background job when you're done with it. --- ## Improving the skill 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-/` 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 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 The description field in SKILL.md frontmatter is the primary mechanism that determines whether the agent 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: ```json [ {"query": "the user prompt", "should_trigger": true}, {"query": "another prompt", "should_trigger": false} ] ``` The queries must be realistic and something a real user of this harness 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_.html`) and open it: `open /tmp/eval_review_.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 **Measure triggering from the session log.** A skill load is an ordinary tool call — `{"type":"tool/call","data":{"name":"skill","arguments":"{\"name\":\"\"}"}}` — recorded in the subagent's persisted session log. Triggering is *observed*, not asked for. Only you can spawn subagents, so you drive the loop and the scripts keep score. Run them as modules from this skill's directory. ```bash python -m scripts.run_loop init loop/ \ --skill-path --eval-set ``` `init` writes `loop/state.json` plus stratified `train-runs.json` and `test-runs.json`, and prints what to do next. Then, per iteration: 1. **Spawn** one subagent per train query (three per query when you want a trigger rate instead of a point estimate). Pass the query verbatim and bound the work so a run does not turn into the full task — "推进到能确定后续做法即可，工具调用不要超过 3 次" is enough. The trigger decision happens before the first tool call, so the bound does not distort it. 2. **Bind** each returned session id: `python -m scripts.run_eval bind loop/train-runs.json --query --session ` 3. **Score:** `python -m scripts.run_eval score loop/train-runs.json --json loop/train-report.json` 4. **Record:** `python -m scripts.run_loop record loop/ --iteration --train-report loop/train-report.json` 5. **Propose only if there are failures.** `python -m scripts.improve_description prompt --report loop/train-report.json --skill-path --out loop/prompt.txt` Read that prompt, answer it yourself, write the `` block to a file, then `python -m scripts.improve_description accept --report loop/train-report.json --skill-path --candidate --out-history loop/history.json --out loop/next.txt`. Put the accepted description into the skill's frontmatter before the next round. 6. **Finish on the held-out split:** repeat steps 1–4 against `loop/test-runs.json` and record it with `--test-report`. `python -m scripts.run_loop best loop/ --out best.txt` prints the highest test score (falling back to train when no test round exists), and `python -m scripts.run_loop report loop/ --out report.html` renders the HTML summary. Running it well: - **All-correct is a result, not a reason to keep going.** No failures means no evidence for a change; rewriting anyway just adds noise you must measure away next round. - **Re-run before trusting a change.** Each query's verdict is a point estimate; treat a rate near the threshold as unresolved and measure it again rather than reading the coin flip. - **The workspace changes the answer.** The same query triggers differently in a repository where the source you would consult is already at hand versus one where it is not. Run the eval in a workspace that resembles the real use case, and write down which workspace and which other skills were installed. - **Catalog composition changes the answer too.** Adding or removing other skills moves individual verdicts, because the description competes with them. ### How skill triggering works Understanding the triggering mechanism helps design better eval queries. Skills appear in the model's available-skills catalog with their name + description, and the model decides whether to load a skill from that description alone. The model only consults skills for tasks it cannot easily handle on its own — a simple one-step query may not trigger a skill even when the description matches perfectly, because the model can just do the work. Complex, multi-step, or specialized queries reliably trigger skills when the description matches. This means your eval queries should be substantive enough that the model would actually benefit from consulting a skill. Simple queries like "read file X" are poor test cases — they will not 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 the `present` tool is available) Check whether you have access to the `present` tool. If you don't, skip this step. If you do, package the skill and present the `.skill` file: ```bash python -m scripts.package_skill ``` After packaging, present the resulting `.skill` file so the user can install it. --- ## Runtimes without subagents The core loop needs subagents: the effect evals spawn one per test case, and trigger evaluation observes one per query. Where they are unavailable, adapt instead of pretending: - **Running test cases**: read the skill's SKILL.md and follow it yourself, one case at a time. This is weaker evidence — you wrote the skill and you are also running it — so lean harder on human review. Skip baseline runs. - **Trigger evaluation**: cannot be done without subagents, because there is nothing to observe. Say that plainly instead of substituting a guess; asking a model to predict whether it would load a skill measures something different, and in measurement it runs optimistic. - **Reviewing results**: with no browser, present the prompt and the output for each case in the conversation, saving any file the user needs to open. Ask for feedback inline: "How does this look? Anything you'd change?" - **Benchmarking and blind comparison**: skip both; they rely on independent runs. - **Packaging**: `package_skill.py` needs only Python and a filesystem. ## Headless / no display If `webbrowser.open()` is unavailable, generate the eval viewer with `--static ` to write a standalone HTML file instead of serving one, and give the user the path. In server mode the "Submit All Reviews" button writes `feedback.json`; in static mode it downloads the same file, so check the downloads folder and copy it into the workspace for the next iteration. ## Updating an existing skill The user might be asking you to update an existing skill rather than create one: - **Preserve the original name.** Keep the directory name and the `name` frontmatter field unchanged. An installed skill named `research-helper` packages as `research-helper.skill`, never `research-helper-v2`. - **Copy to a writable location before editing.** The installed path may be read-only; copy to `/tmp//`, edit there, and package from the copy. - **Stage packaging output in `/tmp/` first**, then copy it where it needs to go. --- ## Reference files The agents/ directory contains instructions for specialized subagents. Read them when you need to spawn the relevant subagent. - `agents/grader.md` — How to evaluate assertions against outputs - `agents/comparator.md` — How to do blind A/B comparison between two outputs - `agents/analyzer.md` — How to analyze why one version beat another The references/ directory has additional documentation: - `references/schemas.md` — JSON structures for evals.json, grading.json, etc. --- Repeating one more time the core loop here for emphasis: - Figure out what the skill is about - Draft or edit the skill - Run subagents that have access to the skill on test prompts - With the user, evaluate the outputs: - Create benchmark.json and run `eval-viewer/generate_review.py` to help the user review them - Run quantitative evals - Repeat until you and the user are satisfied - Package the final skill and return it to the user. Please add steps to your TodoList, if you have such a thing, to make sure you don't forget. Specifically put "Create evals JSON and run `eval-viewer/generate_review.py` so the human can review the test cases" on it, so it actually happens. Good luck!