self-learning

Name: self-learning
Author: Kulaxyz

901stars38forks

How to install self-learning?

kulaxyz/self-learning-skills/self-learning

$npx -y skills add Kulaxyz/self-learning-skills --skill self-learning

Installs into the current project.

Summary

Capture a hard-won "golden path" from the current session as a reusable Agent Skill, so future sessions start already knowing it. Use it (1) right after non-trivial debugging, after working out a multi-step operational workflow, or after rediscovering project facts you didn't kno

Files · 1

View on GitHub

SKILL.md

1	# Self-learning: harvest golden paths into skills
2
3	This skill turns something you just figured out the hard way into a reusable
4	Agent Skill, so the next session — yours or a teammate's — starts already
5	knowing the proven route instead of rediscovering it from scratch.
6
7	It is a meta-skill: it doesn't do the work, it captures how work got done.
8	It's tool-neutral — it works with any agent that reads the Agent Skills format
9	(e.g. Claude Code and Codex, which both load `SKILL.md` skills natively). Where a
10	step differs by tool, the generic version comes first and any tool-specific
11	detail is only an example.
12
13	## Recognize the moment
14
15	Watch for these signals during normal work. Any one of them is a cue to harvest:
16
17	- A task only worked after several attempts, wrong turns, or a correction
18	from the user. The successful path is worth more than the failures around it.
19	- You discovered project-specific facts the agent didn't know up front:
20	where creds/env vars live, which selector or backend talks to a service, a
21	non-obvious command, a required sequence, a gotcha that defies the obvious
22	assumption.
23	- It's an operational workflow likely to recur: reach the dev/prod DB,
24	deploy, run migrations, seed data, verify a change live, run one specific
25	test path, rotate a key, tail the right logs.
26	- The user signals it explicitly: "remember this", "save this as a skill",
27	"don't make me re-explain this next time".
28
29	Act on the cue immediately — don't ask for permission first, whether the
30	user requested it or you noticed it yourself. Harvest the skill, then tell the
31	user what you captured and where (step 5). They can always edit or delete it.
32
33	### Skill, memory, or skip?
34
35	Not every lesson deserves a whole skill — triage first, so you don't bloat the
36	skills list with one-liners:
37
38	- A multi-step, reusable procedure or workflow (how to deploy, reach the DB,
39	run the migration dance, verify live) → harvest it as a skill using the
40	procedure below.
41	- A single standalone fact or one-line correction (an env var name, a path,
42	one gotcha) → if your harness has a lightweight memory/notes facility (e.g. a
43	`MEMORY.md` index), record it there instead; a whole skill is overkill for
44	a one-liner. With no such facility, make a small skill.
45	- A genuinely one-off thing unlikely to recur → skip it.
46
47	When you do harvest, capture the failures too, not just the win: the
48	approaches you ruled out and why often save more time next session than the
49	golden path itself.
50
51	### Promotion rule: don't enshrine guesses
52
53	A skill is authoritative — the next session trusts it without re-deriving it —
54	so hold promotion to a high bar. Only write a skill when all three hold:
55
56	1. A passing check. The path was actually verified — a test passed, the
57	command exited clean, the repro reproduced, the build went green. Record what
58	the check was. "Seemed to work" is not a passing check.
59	2. A named failure pattern. You can name the failure this path avoids or
60	diagnoses (e.g. "stale build cache → phantom type errors"), not a vague
61	"sometimes it breaks".
62	3. At least one ruled-out dead-end. A concrete approach you tried and
63	eliminated, with the reason.
64
65	If any is missing, it isn't a skill yet — leave a tentative note in memory
66	(marked unverified) or skip it. This keeps confident guesses out of the skill
67	set.
68
69	## Harvest procedure
70
71	- [ ] 1. Apply the promotion rule (above). Passing check + named failure
72	pattern + one ruled-out dead-end — or it isn't a skill: note it in memory
73	or skip. Don't proceed on a confident guess.
74	- [ ] 2. Choose scope and name yourself using the heuristics below — don't
75	stop to ask. Default to project scope; pick a clear, specific `name`.
76	- [ ] 3. Dedupe. Look for an existing skill to UPDATE rather than duplicate.
77	List your agent's skills directories — the project

People who liked this also viewed

6 picks

Type

self-learning

901stars38forks

View on GitHub

How to install self-learning?

kulaxyz/self-learning-skills/self-learning

$npx -y skills add Kulaxyz/self-learning-skills --skill self-learning

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Self-learning: harvest golden paths into skills
2
3	This skill turns something you just figured out the hard way into a reusable
4	Agent Skill, so the next session — yours or a teammate's — starts already
5	knowing the proven route instead of rediscovering it from scratch.
6
7	It is a meta-skill: it doesn't do the work, it captures how work got done.
8	It's tool-neutral — it works with any agent that reads the Agent Skills format
9	(e.g. Claude Code and Codex, which both load `SKILL.md` skills natively). Where a
10	step differs by tool, the generic version comes first and any tool-specific
11	detail is only an example.
12
13	## Recognize the moment
14
15	Watch for these signals during normal work. Any one of them is a cue to harvest:
16
17	- A task only worked after several attempts, wrong turns, or a correction
18	from the user. The successful path is worth more than the failures around it.
19	- You discovered project-specific facts the agent didn't know up front:
20	where creds/env vars live, which selector or backend talks to a service, a
21	non-obvious command, a required sequence, a gotcha that defies the obvious
22	assumption.
23	- It's an operational workflow likely to recur: reach the dev/prod DB,
24	deploy, run migrations, seed data, verify a change live, run one specific
25	test path, rotate a key, tail the right logs.
26	- The user signals it explicitly: "remember this", "save this as a skill",
27	"don't make me re-explain this next time".
28
29	Act on the cue immediately — don't ask for permission first, whether the
30	user requested it or you noticed it yourself. Harvest the skill, then tell the
31	user what you captured and where (step 5). They can always edit or delete it.
32
33	### Skill, memory, or skip?
34
35	Not every lesson deserves a whole skill — triage first, so you don't bloat the
36	skills list with one-liners:
37
38	- A multi-step, reusable procedure or workflow (how to deploy, reach the DB,
39	run the migration dance, verify live) → harvest it as a skill using the
40	procedure below.
41	- A single standalone fact or one-line correction (an env var name, a path,
42	one gotcha) → if your harness has a lightweight memory/notes facility (e.g. a
43	`MEMORY.md` index), record it there instead; a whole skill is overkill for
44	a one-liner. With no such facility, make a small skill.
45	- A genuinely one-off thing unlikely to recur → skip it.
46
47	When you do harvest, capture the failures too, not just the win: the
48	approaches you ruled out and why often save more time next session than the
49	golden path itself.
50
51	### Promotion rule: don't enshrine guesses
52
53	A skill is authoritative — the next session trusts it without re-deriving it —
54	so hold promotion to a high bar. Only write a skill when all three hold:
55
56	1. A passing check. The path was actually verified — a test passed, the
57	command exited clean, the repro reproduced, the build went green. Record what
58	the check was. "Seemed to work" is not a passing check.
59	2. A named failure pattern. You can name the failure this path avoids or
60	diagnoses (e.g. "stale build cache → phantom type errors"), not a vague
61	"sometimes it breaks".
62	3. At least one ruled-out dead-end. A concrete approach you tried and
63	eliminated, with the reason.
64
65	If any is missing, it isn't a skill yet — leave a tentative note in memory
66	(marked unverified) or skip it. This keeps confident guesses out of the skill
67	set.
68
69	## Harvest procedure
70
71	- [ ] 1. Apply the promotion rule (above). Passing check + named failure
72	pattern + one ruled-out dead-end — or it isn't a skill: note it in memory
73	or skip. Don't proceed on a confident guess.
74	- [ ] 2. Choose scope and name yourself using the heuristics below — don't
75	stop to ask. Default to project scope; pick a clear, specific `name`.
76	- [ ] 3. Dedupe. Look for an existing skill to UPDATE rather than duplicate.
77	List your agent's skills directories — the project

People who liked this also viewed

6 picks

Type