Skill coach
A starter kit for growing your own agent skills instead of borrowing someone else's. For Claude Code and Cursor.
npx -y skills add flll/grow-your-own-skills --skill skill-coachAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when the user corrects you about the same thing for the second time, or when they ask to create, improve, or prune a skill. Also use when you catch yourself repeating a mistake an existing skill was meant to prevent. Not for one-off fixes that will never recur.
SKILL.md
1.9 KB, 391 tokens by cl100k_base, as published. Nobody here has run it
skill-coach — turn repeated corrections into skills
The two-strike rule
The second time the user gives you the same correction, stop and offer:
"That's the second time — shall I turn this into a skill?" If yes, draft one
from _template/SKILL.md and show the draft before saving it.
Writing rules
- The description decides whether the skill ever gets read. Start it with "Use when..." and list concrete trigger situations — not what the skill does.
- One skill, one concept. If a draft covers two ideas, split it.
- Budget: core stays under ~200 words; any other skill under ~500. Move long reference material into extra files inside the skill folder and link them.
- Write instructions, not essays. Prefer "do X" plus one line of why over an all-caps "NEVER" — a rule the model understands beats a rule it must obey.
- Do not create a skill for: a one-off fix, general knowledge the model already has, or anything a linter / CI check could enforce mechanically.
- Base every rule on a failure that actually happened (twice). No speculative rules for imagined failures.
After creating a skill
- Add one row to the routing table in
core/SKILL.md. - Watch the next sessions: if the skill fired but the mistake still happened, tighten the wording. If it never fires, rewrite the description.
Monthly pruning
Once a month, delete or merge any skill that didn't get used. Growing a skill garden means pruning, not accumulating.
Credits: distilled from anthropics/skills (skill-creator) and obra/superpowers (writing-skills).
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most learn study skills give in 391 tokens
Counted across 546 of the 573 authors here whose files we hold, read 2026-08-07
- Calculate the zone of proximal development before teachingin 25 of 546, across 8 files
- Produce self-contained HTML lessonsin 24 of 546, across 8 files
- Record user preferences in a notes filein 23 of 546, across 5 files
- Maintain a teaching workspace in the current directoryin 21 of 546, across 4 files
- Find high-quality resources before writing lessonsin 19 of 546, across 5 files
- Make lessons beautiful, short, and quickly completablein 19 of 546, across 3 files
- Create reusable components for lessonsin 19 of 546, across 5 files
- Create compressed reference documents for quick lookupin 19 of 546, across 3 files
- Update the mission file and records upon mission changesin 16 of 546, across 2 files
- Set min_dist to 0.0 for clustering preprocessingin 16 of 546, across 6 files
- Populate the mission file before teachingin 15 of 546, across 1 file
- Include interactive feedback loops in lessonsin 15 of 546, across 1 file
Said here and by no other author read
- offer to create a skill after the second identical correction
- draft a skill from the template
- show the draft before saving it
- start skill descriptions with use when and concrete triggers
- limit each skill to one concept
- split drafts covering two ideas
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.