Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →tsinghua-fib-lab/my-custom-skill
Example custom agent skill — a template to get started.
Passed 15 of 19 test cases with the skill, and 10 of 19 without it.
and 3 more in the full report
6 fixed11 passed either way5 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
When appending structured log entries for the daily activity record, what specific relative file path is targeted?
| Checked | Without skill | With skill |
|---|---|---|
| Targets path journal.jsonl for log storage | Missed | Passed |
The instructions your model reads when the skill loads.
This is a template for creating a custom agent skill.
There are two valid skill patterns in the current architecture:
No script field in frontmatter. This SKILL.md is injected into the LLM's context
when activate_skill is called. The LLM then uses built-in atomic tools
(read, write, append, list, grep, ask_env) to accomplish the task.
This is the primary extension mechanism — like Claude Code's slash commands.
Add script: scripts/my-script.py to frontmatter. The script is executed as a
cached in-process module via entrypoint(argv, ctx) when available, with dynamic
wrapper/subprocess fallback. It should communicate via the returned stdout string
and file I/O under ctx.workspace_root. Scripts cannot access the LLM or
environment router directly — use this only for deterministic computation.
When this skill is activated:
ask_env to query the environment for relevant information.read / write / append to persist state.execute_skill_script for deterministic computation when a script exists.finish with a summary when finished.When activated, the agent should:
ask_env with instruction: "What happened recently? Summarize recent events."read path journal.jsonl to load previous entries if it exists.append path journal.jsonl to add today's entry.finish with summary of what was journaled.test cases passed, vs 10 / 19 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.