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 skills add cathrynlavery/agent-improvement-loop --skill learn-loopgit clone --depth 1 https://github.com/cathrynlavery/agent-improvement-loopWrote 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/skills/cathrynlavery/agent-improvement-loop/learn-loop)<a href="https://agentmods.dev/skills/cathrynlavery/agent-improvement-loop/learn-loop"><img src="https://agentmods.dev/badge/skills/cathrynlavery/agent-improvement-loop/learn-loop/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/cathrynlavery/agent-improvement-loop/learn-loop"><img src="https://agentmods.dev/badge/skills/cathrynlavery/agent-improvement-loop/learn-loop.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00063 | $0.01346 |
| Opus 5.5 | $0.00025 | $0.00538 |
| Sonnet 5.5 | $0.00013 | $0.00269 |
| Haiku 4.5 | $0.00006 | $0.00135 |
Grade A, and why
learn-loop 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 1mo 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 — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Learn Loop
Turn captured friction into verified system improvements:
survey → select → approve → fix → verify → decide
The canonical store is ${AGENT_LEARNINGS_ROOT:-~/.agents/learnings}. Machine-owned files under entries/ are immutable evidence. Effective outcomes live in leader-owned decisions/; ID corrections live in rekeys/.
Modes
- Survey — recommend what to fix without changing code or store outcomes.
- Fix — execute an approved bounded batch through verification and decision.
- Reconcile — resolve divergent-copy conflicts. Read STORE-PROTOCOL.md before any store mutation.
- Resume — continue an existing plan or PR from current state; do not repeat completed work.
1. Pin the snapshot
- Read
config.jsonand runlearnings status. - Use
ACTIVE.md,catalog.json, andconflicts.jsonon the leader; on a writer, usefleet/for survey and move mutations to the leader. - Stop store mutation when generated views disagree or the catalog reports invalid files, stale acknowledgements, rekey errors, decision errors, or integrity errors.
- Read the fix queue and prior
FIX-PLAN-*-RESULTS.mdfiles. Re-check live PRs and external dependencies instead of trusting their recorded status. - When the request concerns loop health, check recent harvest, collection, and triage run lines plus the leader's unattended runtime.
Snapshot complete when: one healthy catalog generation is pinned and every previously touched candidate is classified as finished, still active, or superseded.
2. Pass the evidence gate
For each candidate, read every referenced evidence copy and inspect the current target. A title or aggregate row is a pointer, not proof.
Assign exactly one disposition:
- Promote — encode a durable rule in an existing instruction, skill, test, hook, or wrapper.
- Code fix — change one repository with a named test and shipping target.
- Operational fix — repair a scheduler, configuration, migration, or live integration.
- Reconcile — acknowledge compatible copies or rekey unrelated incidents.
- Close without change — duplicate, stale, already fixed, one-off, or intentionally declined.
- Blocked — requires credentials, payment, destructive authority, merge approval, or unresolved human judgment.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 1mo ago First seen · 116 lines · 63 tokens per session scan A 7a670bd2dd0b
learn-loop is a skill published in the GitHub repository cathrynlavery/agent-improvement-loop (70 stars, last pushed 17d ago), licensed MIT. It adds 63 tokens to every session and 1,346 once invoked, about $0.0003 per session on Opus 5.5. 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 skills, from other repositories
webgl-holographic-foil
A self-contained WebGL2 hero: thin-film interference over a crushed-foil surface whose palette shifts with the viewing angle; move the cursor to tilt the film.
ligandmpnn
Inverse-fold a backbone with ligand, nucleic-acid, and metal context using LigandMPNN (Dauparas et al. 2023, github.com/dauparas/LigandMPNN). Reach for this skill to redesign the residues lining a binding pocket around a bound small molecule or cofactor, to design metal-coordinating sites where the geometry must be…
evo2
Score, embed, and generate DNA sequences with Evo 2, a long-context genomic foundation model. Use this skill when: (1) Computing per-nucleotide or per-sequence likelihoods for variant effect scoring, (2) Embedding genomic windows for downstream classification, (3) Generating DNA conditioned on a prefix, (4) Scoring…
bo-process-optimization
A method for examining a business workflow to find waiting, rework, handoff waste, and other delays, then proposing improvements and ways to measure them. It separates observed facts from possible explanations and suggestions.
starcat-ui-prototype
A design skill for creating macOS interface prototypes and visual designs for Starcat, a GitHub star-management and AI knowledge-organising app. It follows Apple’s Liquid Glass design language for macOS 26.
llm-wiki
Build and maintain a workspace-local, Obsidian-friendly research vault using Karpathy's LLM Wiki pattern.