Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/tacoda/keystone/learngit clone --depth 1 https://github.com/tacoda/keystoneWhat 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 | $0.00016 | $0.00555 |
| Opus 5 | $0.00008 | $0.00278 |
| Sonnet 5 | $0.00003 | $0.00111 |
| Haiku 4.5 | $0.00002 | $0.00056 |
Grade A, and why
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 2d ago.
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 — 57 lines — stays where its author put it; the contents beside it link to each section on GitHub.
learn
Capture a learning candidate from a surprise, an incident, or a review finding. Writes to .charter/learning/inbox/ for later synthesis. Read .charter/learning/README.md.
Activities
Write a candidate to .charter/learning/inbox/<timestamp>-<slug>.md where:
<timestamp>isYYYY-MM-DD-HHMMin UTC<slug>is a short kebab-case description of the insight
The candidate file uses this shape:
---
captured: <ISO date>
source: <what triggered this — review finding, surprise during implementation, etc.>
proposed-layer: <corpus/principles | guides/idioms/<stack> | guides/process | sensor>
proposed-globs: # optional; see below
- "src/billing/**"
- "tests/billing/**"
---
## What happened
<concrete observation — code, output, or interaction>
## Why it matters
<the principle, idiom, or rule this implies>
## Proposed change
<the smallest charter edit that would prevent the next incident>
proposed-globs: — record the paths the lesson came from
When the surprise happened in a specific region of the codebase, list the paths in proposed-globs:. These are the touched files (or their parent directories' globs) from the interaction that produced the candidate. Synthesize uses this as signal when deciding the guide's actual globs:.
- If the lesson is cross-cutting (would apply to any file in any stack), omit the field. Synthesize will default to no
globs:on the promoted guide. - If the lesson is regional, list the patterns that match the affected files. Prefer existing region globs from
corpus/state/CODEBASE_STATE.mdover hand-invented patterns — globs should reflect the real codebase, not the abstraction the rule is about. - Never set
proposed-globs:wider than where the surprise actually occurred. Synthesize narrows on user confirmation; widening at learn-time loses the evidence.
Gate
Learn writes only to the inbox. Promotion into corpus / guides happens in synthesize, not here. Capture liberally; prune aggressively later.
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.
- 2d ago First seen · 57 lines · 16 tokens per session scan A 1730469b6500
learn is a command published in the GitHub repository tacoda/keystone (44 stars, last pushed 2mo ago), licensed MIT. It adds 16 tokens to every session and 555 once invoked, about $0.0001 per session on Opus 5. 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 2026-08-30.
Other commands, from other repositories
learn
Force claude-smart to extract learnings from this session now.
component
Scaffold a new React component grounded in the paper-mono primitives. Requires explicit kind or a nearest-existing-component match. No empty divs, no speculative scaffolding.
memory-store
Store an insight, decision, or pattern to memory.
review
Cold re-quiz on code that already shipped — your own session commits, not the change in front of you.
no-vibe
Enter no-vibe mode in OpenCode (tutor mode, no direct project file writes).
teach-me-testing
Teach testing progressively through structured sessions. Use when user says ""lets learn testing"" or ""I want to study test practices"".