When a test fails just before your commit, your debugging method determines whether the next chat investigates or guesses. A skill is a folder of instructions the model opens when it decides it needs them. Of 11,549 debugging skills, almost all load before you ask for help. Occasional use can therefore cost more than expected.
When a debugging skill reaches the session
A skill’s name and description enter the context first. The context is everything the model can see while it answers you. When the model chooses the skill, its whole body enters the context for that session.
Only 313 skills wait until you call them. The other 11,236 load every session, even when your problem is unrelated. A glob is a filename pattern, like *.tsx. It can narrow files, but it does not decide when a skill opens.
What a debugging skill costs while idle
Across always-loaded debugging skills, the median is 55 tokens every session. A token is the unit your AI plan bills by, about three quarters of a word. One in ten costs 130 tokens or more.
The largest always-loaded example is incident-investigate, at 771 tokens and 1,317 lines. You pay that cost before the bug becomes relevant.
If you open 20 sessions a day, the 55-token median becomes 1,100 tokens across a day. An on-demand method costs a median 864 tokens when loaded, but nothing while waiting.
What the safety scan says about debugging skills
The scan gave 10,924 skills grade A, 338 grade B, and 287 grade C or D. Grade A means the scan found nothing it treats as risky. It does not mean the instructions are automatically right for your project.
Making network calls was the most common finding, appearing 896 times. The scan also found 188 requests for root access and 180 recursive force deletes.
The original is the first copy of a file we saw; everything matching it afterwards is a copy. Three examples show why the line matters: rest-graphql-debug has grade D with external data and remote-code findings. customizations-in-the-agent-host has grade C for recursive force deletion. issue-triage has the same grade and finding.
Each finding prints the file line that triggered it. Inspect that line before deciding whether to use the method.
Five debugging skills to try first
“Every session” means the instructions load before the bug appears.
- troubleshoot: Reads JSONL debug logs to explain slow requests, missing tools, or unexpected agent choices. 55 tokens every session, 272 lines, original, 1 copy.
- add-or-fix-type-checking: Repairs Python type-checking errors from local checks, CI, or pull-request logs. 41 tokens every session, 216 lines, original, 0 copies.
- node-compat: Runs a Deno compatibility test, traces failures, and decides whether to fix or classify them. 37 tokens every session, 188 lines, original, 0 copies.
- code_explorer: Locates source files, connected interface components, and tests related to a bug. 28 tokens every session, 34 lines, original, 0 copies.
- investigate: Investigates evidence and tests possible causes before implementing a fix. 15 tokens every session, 729 lines, original, 0 copies.
When a hook beats a debugging skill
A hook is a shell command your agent runs for you when something happens. Choose a hook when you need an automatic action after a file change, before a commit, or after a test finishes.
A debugging skill cannot fire on an event. It waits for the model to decide it is relevant, although it may tell the model to run a script. Use the skill to reason about a bug, and a hook to enforce a repeatable event.
Questions people ask
Will this work in Cursor? Of these, 95 are written for Cursor and 10,729 for Claude Code. Check the target agent before installing one.
Is grade A proof that a skill is safe? No. Review any method that downloads and executes remote code, a finding recorded 56 times. Read the flagged line before deciding.
Is loading a long skill once a problem? A one-off load can fit inside the current models’ 200,000-token context window. If you expect repeated use, choose a method that waits for a match.
Do it now
Run npx agentmods cost in your project to see what the mods you already have cost per session, or browse the skills catalogue.