Borrowing it
Nothing to install: this file belongs to yash-gadodia/claude-init. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/yash-gadodia/claude-init/main/.github/prompts/self-learn.mdgit clone --depth 1 https://github.com/yash-gadodia/claude-initWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/yash-gadodia/claude-init/self-learn)<a href="https://agentmods.dev/commands/yash-gadodia/claude-init/self-learn"><img src="https://agentmods.dev/badge/commands/yash-gadodia/claude-init/self-learn.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00000 | $0.00627 |
| Opus 5 | $0.00000 | $0.00313 |
| Sonnet 5 | $0.00000 | $0.00125 |
| Haiku 4.5 | $0.00000 | $0.00063 |
Grade A, and why
self-learn scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured today.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 46 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Weekly Self-Learning Loop
Study the Claude Code ecosystem and propose AT MOST ONE concrete improvement to claude-init per run. Evidence over speculation. If nothing clears the bar this week, that is a valid outcome.
Phase 1: Gather
- Read
LEARNINGS.mdFIRST — the reject list and last-checked dates. Never re-propose anything marked rejected unless its revisit date has passed. - Check upstream changes since the
Last checkeddates in LEARNINGS.md:anthropics/claude-codeCHANGELOG.md — new/changed Claude Code featuresanthropics/skills— new official Agent Skills or format changes- Official docs at code.claude.com/docs (skills, hooks, sub-agents, plugins, memory pages)
- Scan comparable repos for new patterns (recent commits/README changes):
obra/superpowers— skill methodologytrailofbits/claude-code-config— security-first confighesreallyhim/awesome-claude-code— new ecosystem entries worth studyingVoltAgent/awesome-agent-skills— cross-platform skill patterns
Phase 2: Analyze
For each candidate idea, require ALL of:
- Makes claude-init's GENERATED configs better (not just interesting)
- Fits in templates/, docs, or the four self-skills — no core rewrite
- Has a citable source (URL to repo, changelog entry, or doc page)
- Not in LEARNINGS.md as rejected (or revisit date has passed)
Pick the single strongest idea, or none.
Phase 3: Apply (commit directly to main)
If you have one improvement:
- Make the change on
main— max ~200 changed lines, templates/docs/skills only - Update
LEARNINGS.md: add this cycle's entry (adopted idea + any rejected candidates with reasons), refresh allLast checkeddates - Run
bash tests/run.sh— MUST pass before you commit. If it fails and you can't fix it within the diff budget, revert everything except the LEARNINGS.md entry (log the idea as "attempted, tests failed") and commit only that - Commit to main with the source URLs in the message, prefixed
[self-learn], and push
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- today First seen · 46 lines · 0 tokens per session scan A c79e3ca8298f
self-learn is a command published in the GitHub repository yash-gadodia/claude-init (5 stars, last pushed yesterday), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 627 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-08.
Other commands, from other repositories
analytics-review
Review analytics and recent commits, then create tasks.
fix-issues
Pick up one approved task, implement the fix, and close it.
architecture-review
Architecture + code-health audit — module boundaries, then fallow on TS/JS — and a task per finding.
check-ux
Walk the key flows in a real browser at mobile and desktop viewports, then a task per friction point.
docs-audit
Audit docs/ against the code and recent history; file a task for each significant drift.
submit-to-directories
Submit the product to the directory catalog in /.darkflow/directories.csv through a real browser, and track what has already been submitted.