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 megamen32/LastHumanCommit --skill mini-improvement-loopgit clone --depth 1 https://github.com/megamen32/LastHumanCommitWrote 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/megamen32/lasthumancommit/mini-improvement-loop)<a href="https://agentmods.dev/skills/megamen32/lasthumancommit/mini-improvement-loop"><img src="https://agentmods.dev/badge/skills/megamen32/lasthumancommit/mini-improvement-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/megamen32/lasthumancommit/mini-improvement-loop"><img src="https://agentmods.dev/badge/skills/megamen32/lasthumancommit/mini-improvement-loop.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00058 | $0.00776 |
| Opus 5.5 | $0.00023 | $0.00310 |
| Sonnet 5.5 | $0.00012 | $0.00155 |
| Haiku 4.5 | $0.00006 | $0.00078 |
Grade A, and why
mini-improvement-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 — 58 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Mini improvement loop
One named cycle for closing a concrete defect list (CRITICAL/HIGH/MEDIUM/LOW findings from a review, audit, incident, or test round). Not for open-ended design work — that is Full. Not for a single obvious bug — fix it directly.
The loop
- Fix via workers in parallel. Give each worker one bounded slice with a self-contained contract: absolute paths, the exact findings, invariants that must survive, and required test evidence (scenarios with expected exit codes and filesystem effects). Forbid git operations in the worker contract — parallel workers fight over one index; the Lead integrates.
- Lead integrates. Review every diff personally, add only cross-cutting glue, commit small and correct (each logical fix its own commit), push.
- Critique round. Two independent subagents in parallel: an adversarial Reviewer that must verify every claimed fix is real AND hunt for new bugs the fixes introduced, plus a Tester that re-derives scenarios in a sandbox without trusting prior test results or prior test drivers.
- Triage the findings. CRITICAL/HIGH → fix now. MEDIUM → fix if each is under ~10 minutes. LOW → fix the one-liners in the same pass, consciously drop the rest. Small fixes are the Lead's direct work; only a slice needing its own investigation goes back to a worker.
- Re-critique only the delta. One focused reviewer confirming finding→fix mapping and scanning the delta for regressions is enough. Repeat until a round returns no CRITICAL/HIGH. Default cap: three critique rounds; the third is a delta confirmation, not a full re-audit.
- Close on the real surface. One real run or deployment of the changed thing with its output in the final report (exit code, state files created, canary log lines). The loop ends on that proof, not on "fixes committed".
Convergence rules
- Every round's fixes are their own commits, so any round is revertible and the history reads as the loop it was.
- A finding that survives two fix attempts is a design decision, not a bug: stop fixing, present two compressed variants to the user (AskHuman when away).
- Sandbox discipline: workers and testers never touch real user-data paths — sed-copied scripts, env-overridden homes, temp trees. The Lead's single real run at the end is the only live fire.
- A reviewer or tester that could not execute (no shell, no write) must say so in the report; their verdict is then code-read only and cannot replace the real-surface step.
- Worker bugs happen too: when a fix's own test exposes a bug in the fix (tautological check, lost default branch), it is fixed in the same loop, not hidden — the loop converges on truth, not on green.
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 · 58 lines · 58 tokens per session scan A e920858a42b6
mini-improvement-loop is a skill published in the GitHub repository megamen32/LastHumanCommit (2 stars, last pushed yesterday), licensed MIT. It adds 58 tokens to every session and 776 once invoked, about $0.0002 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 2026-09-07.
Other skills, from other repositories
systematic-debugging
4-phase root cause debugging: understand bugs before fixing.
langsmith-observability
LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitoring production systems, or building systematic testing pipelines for AI applications.
test-failure-explorer
Import and query an existing Vitest 3.2 JSON report to explore test failures, file-level errors and original diagnostic evidence.
experimental-code-coverage-local-debugger
Runs code coverage locally via Universal Test Runner (UTR) or helper scripts, mimicking LUCI trybots. Activate when CQ tryjobs fail or underreport coverage, to test local GN/recipe repairs before uploading, or to debug hermetic crashes.
verify-omnigent
Drive Omnigent the way a user does and prove a behavior with recorded evidence, using an isolated server, runner, and mock model plus a feature map of every user entry point. Load before reproducing a user-facing bug, before claiming a fix works, or when reviewing whether a change covered every surface (session vs.…
cli-e2e
Write, modify, or debug Docker-based Composio CLI end-to-end tests under ts/e2e-tests/cli, including binary invocation, fixture isolation, output assertions, and package manifests. Use for CLI E2E test suites only; use cli-command for CLI source implementation.