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/nel-neru/libraium/improvegit clone --depth 1 https://github.com/nel-neru/LibrAIumWhat 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 | $0.00026 | $0.01132 |
| Opus 5 | $0.00013 | $0.00566 |
| Sonnet 5 | $0.00005 | $0.00226 |
| Haiku 4.5 | $0.00003 | $0.00113 |
Grade A, and why
improve 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 yesterday.
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 — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Run exactly one iteration of LibrAIum's improvement loop. One iteration = one focused improvement, verified and committed. Do not batch multiple unrelated improvements into a single iteration.
1. Load state
Read .claude/improvement-backlog.md.
If it does not exist (first iteration): seed it by auditing the repo along these dimensions, then pick the top item and continue. Write concrete, single-commit-sized items — not vague themes.
- Robustness (Rust) —
unwrap()/expect()/panic paths insrc-tauri/src/, unhandled I/O errors, edge cases in store/search/github/gitops (empty data dir, malformed entry mid-library, concurrent refresh, git repo in weird states) - Robustness (MCP server) — input validation gaps, unhelpful error messages, behavior on corrupt/missing data files
- Test coverage — untested branches in Rust modules; the MCP server has only the smoke test — targeted unit tests for
lib/store.js/lib/suggest.js - Dual-format parity — adversarial edge cases (unicode slugs, CRLF, empty body, frontmatter-like body content, huge tags) → new fixtures in
tests/fixtures/format/proving both sides agree - Frontend (Svelte 5) — empty/loading/error states, race conditions on rapid navigation, keyboard access, feedback on slow network ops
- DX / docs — README accuracy, script failure messages, CLAUDE.md drift
- CI — speed, cache hit rate, flakiness
Backlog format:
# Improvement Backlog
## Pending (ordered: highest value first)
- [ ] P1 <one-line item> — <why it matters, 1 sentence>
## Done
- [x] <item> — <commit hash> (<date>)
## Rejected
- <item> — <why not worth doing>
Priority order: P1 correctness & data-safety bugs → P2 missing tests for existing behavior → P3 error handling & user-facing failure messages → P4 dual-format hardening (new fixtures) → P5 docs/DX/CI polish.
2. Pick ONE item
Take the highest-priority pending item that fits in a single focused session. If the top item is too big, split it into smaller backlog entries and take the first piece. State clearly which item you picked and why.
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.
- yesterday First seen · 78 lines · 26 tokens per session scan A 502a4afa2dab
improve is a command published in the GitHub repository nel-neru/LibrAIum (0 stars, last pushed 5d ago), licensed MIT. It adds 26 tokens to every session and 1,132 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 2026-08-31.
Other commands, from other repositories
codebase-audit
Survey a codebase as a senior advisor and produce prioritized, self-contained improvement plans. Read-only on source code. Use standalone before iteration-start to discover what's worth doing, or independently to build a prioritized backlog.
memory-update
Run the Memory Update protocol. Checks whether decisions, questions, assumptions, risks, tasks, or opportunities emerged in the current task and writes them to the right files, then prints the required end-of-task summary.
reconcile
Clear the decision review backlog. Walk the decisions that are due for review and, for each, decide together: still true, needs evidence, changed, or archive.
queue
"What's coming up — plan steps, releases, backlog, and overdue decisions?".
status
Show wiki state at a glance — counts, recent activity, index drift, inbox backlog, collectionPurpose coverage, Core Context snapshot age.
decompose
Feature/story breakdown — consume a ready-for-decompose scope doc and produce a task graph in the issue tracker. Re-entrant; tracker-as-canonical with vault breadcrumb.