Skill Ledger

Keep the SKILL.md files you use in one place. Skill Ledger checks each one against the Agent Skills spec, adds up what your skill names and descriptions cost in context and compares it with Claude Code’s skill-listing budget (1% of the context window), flags skills whose descriptions compete, and re-runs your should-trigger and should-not-trigger prompts whenever you edit a description.

How it works

1. Add skills: paste a SKILL.md, pick files, or pick your whole skills folder. They are read in your browser.
2. Open the Library. Skills with something to fix come first; each lists its problems with the line number, from a missing or oversized name to a description that YAML cannot parse.
3. Write a few test requests per skill (“should trigger” and “should not trigger”). When you rewrite a description, or add a new skill that steals a request, the tests show it straight away.
4. Open Budget and overlap to see how big your skill listing is against Claude Code’s budget of 1% of the context window (above it, Claude Code drops descriptions), and which pairs of skills compete.

What it checks

Against the open Agent Skills specification: the name (1 to 64 characters, lowercase letters, digits and single hyphens, matching the directory), the description (required, at most 1,024 characters), compatibility (at most 500), metadata values that are strings, and a body that is not too long. Against Claude Code’s skills page: its extra fields are recognised, and the combined description and when_to_use text (the name is not counted) is flagged where Claude Code’s listing cuts it off at 1,536 characters. Claude Code also keeps its whole skill listing within 1% of the context window and drops descriptions beyond that, least-used skills first; the Budget tab shows where your library stands. It also catches the quiet one: an unquoted description containing a colon and a space, which strict YAML parsers reject.

What it does and does not do

Token counts are estimates at four characters per token; real tokenizers differ. The trigger lab and the trigger tests compare words: they drop everyday words, trim plural endings, and weight words few of your skills use. That finds vague descriptions, missing keywords and near-duplicates. It is not a simulation of any agent, which chooses with a language model, so a passing test is a good sign and not a guarantee. Only SKILL.md is read; scripts and reference files in a skill folder are ignored. Each SKILL.md may be up to 16,000 characters.

Free and paid

Everything above is free for 12 skills (the last 3 saved versions of each) and runs in your browser; nothing is sent anywhere. The paid plan is $5 a month: up to 150 skills with 20 saved versions each, synced between your devices with a key, with no account or password. You can always download a backup. Payments are handled by Stripe; the refund policy is linked in the footer.