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 AlexShchuka/neuro-matrix --skill harness-improvegit clone --depth 1 https://github.com/AlexShchuka/neuro-matrixWrote 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/alexshchuka/neuro-matrix/harness-improve)<a href="https://agentmods.dev/skills/alexshchuka/neuro-matrix/harness-improve"><img src="https://agentmods.dev/badge/skills/alexshchuka/neuro-matrix/harness-improve/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/alexshchuka/neuro-matrix/harness-improve"><img src="https://agentmods.dev/badge/skills/alexshchuka/neuro-matrix/harness-improve.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 | $0.00074 | $0.00948 |
| Opus 5 | $0.00037 | $0.00474 |
| Sonnet 5 | $0.00015 | $0.00190 |
| Haiku 4.5 | $0.00007 | $0.00095 |
Grade A, and why
harness-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 11d 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 — 54 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Harness-improve: changing the protocol without meta-neuroslop
BLUF: improvements to the harness are the highest-slop-risk artifact in this repo — plausible protocol text detached from incidents is neuroslop about neuroslop. Entry requires a live anchor; every change class has its own gate; a null measurement result is information, not failure.
Stage 0 — anchor or stop
Admissible anchors, in order of strength:
- A reproduced incident — tool output from this or a documented session.
- A structured harness-gap report (
.github/ISSUE_TEMPLATE/harness-gap.yml). - A measurement — an eval run, an A/B result, a counted pattern across sessions.
- An explicit owner decision.
«Would be nice» without an anchor → record as a ROADMAP row candidate or drop. Do not proceed to implementation on an unanchored idea: critique of the harness demands the same evidence bar as critique of code.
Stage 1 — verify against current main
- Pull main; confirm the gap still exists by reading the current files, not the report. Reports go stale: the 2026-06-11 issue revision found a «missing» acceptance kit that had been merged before the issue was written.
- Check
ROADMAP.mdand open issues for an existing row or tracker. A recorded decision (ADR, ROADMAP status, protocol doc) must be quoted before re-opening, and re-opening requires new evidence — not repeated association.
Stage 2 — classify the change and pick its gate
| Class | Typical diff | Gate before merge |
|---|---|---|
| Mechanical | hooks/, scripts/ |
bash -n + python3 -m json.tool on JSON + the matching scripts/selftest_*.sh (extend it with the new behavior — a test of a hand-copied reimplementation is decorative coverage) |
| Behavioral text | invariants.txt, agents/, references/, CLAUDE.md |
single-line invariant format (selftest_random_invariant.sh); role-subset inheritance ONLY via the eval-gated N8 path (ROADMAP) |
| Calibration | eval/criteria.md, judge prompts, role subsets, agent system prompts |
eval run (label run-eval); reconcile every consumer of changed counts/contracts (ci_eval.py, docs — the 16-vs-17 drift class); expect judge noise — report null results as-is |
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.
- 11d ago First seen · 54 lines · 74 tokens per session scan A cfeae61de5c8
harness-improve is a skill published in the GitHub repository AlexShchuka/neuro-matrix (3 stars, last pushed 2mo ago), licensed MIT. It adds 74 tokens to every session and 948 once invoked, about $0.0004 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 skills, from other repositories
test-driven-development
Forces the strict Red-Green-Refactor development cycle. Requires one failing behavior test through a public interface before each implementation slice.
release-checklist
Generates a comprehensive pre-release validation checklist for software versions covering build verification, deployment readiness, and release risk. Use when preparing to release a product version or when the user mentions release checklist or pre-release validation.
review-spec
Reviews a product, technical, API, UI, or implementation spec for completeness, testability, architectural fit, and readiness before planning or implementation.
verification-before-completion
Requires fresh verification evidence before claiming work is complete, fixed, passing, safe, ready, merged, or clean. Use before completion claims, commits, PRs, merge readiness, or moving to the next task.
verify-qa
A conditional end-to-end quality-assurance workflow for changes to a user interface. End-to-end testing checks a feature through the whole user path, often in a real browser, and this workflow also includes performance and accessibility checks.
harness-engineering-guide
Audit, design, and implement AI agent harnesses for any codebase. A harness is the constraints, feedback loops, and verification systems surrounding AI coding agents — improving it is the highest-leverage way to improve AI code quality. Three modes: Audit (scorecard), Implement (set up components), Design (full…