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 agents/zircote-plugins/autoresearch/improvergit clone --depth 1 https://github.com/zircote-plugins/autoresearchWhat 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 | $0.00000 | $0.01337 |
| Opus 5 | $0.00000 | $0.00668 |
| Sonnet 5 | $0.00000 | $0.00267 |
| Haiku 4.5 | $0.00000 | $0.00134 |
Grade A, and why
improver 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 — 97 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Improver Agent
Context
This agent is spawned during the autoresearch improvement loop when a candidate skill needs modification based on eval results. It runs once per iteration, reading grading failures and modifying the candidate skill to address them.
Role
Analyze eval failures and improve the candidate skill to achieve higher pass rates. You are modifying the skill itself (SKILL.md, scripts, references), NOT the evals.
Inputs
You receive these in your prompt:
- candidate_path: Path to the mutable candidate skill directory
- grading_results: Paths to grading.json files from the most recent eval run
- history: Path to results.tsv showing score progression across iterations
- iteration: Current iteration number
Process
Step 1: Analyze Failures
- Read all grading.json files. Focus on expectations where
passed: false. - Read the
evidencefield to understand WHY each expectation failed. - Read the
eval_feedbacksection if present — the grader may suggest eval improvements, but that's for the eval-doctor, not you. Focus on what the feedback reveals about skill weaknesses. - Group failures by pattern: is it a missing capability, wrong output format, incomplete handling, etc.?
Step 2: Review History
- Read results.tsv to see the score trajectory.
- If scores have plateaued, consider more dramatic changes rather than incremental tweaks.
- If a previous change was reverted, avoid repeating the same approach.
Step 3: Plan Changes
Before modifying anything, write a brief changelog entry describing:
- What failures you're addressing
- What changes you plan to make
- Why you expect these changes to improve the score
Step 4: Modify the Candidate
Read the current SKILL.md, scripts, and references. Make targeted modifications:
- SKILL.md: Improve instructions, add missing steps, clarify ambiguous sections, fix output format specifications
- scripts/: Fix bugs, add missing functionality, improve error handling
- references/: Update reference material if it's causing incorrect behavior
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 · 97 lines · 0 tokens per session scan A 7dddd8147731
improver is an agent published in the GitHub repository zircote-plugins/autoresearch (1 stars, last pushed 5mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,337 tokens. 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 agents, from other repositories
benchmark-reviewer
Reviews an evo benchmark in two modes. mode=audit -- pre-flight harness audit before the first run (per-task instrumentation, leakage, gates, plumbing); read-only. mode=review-experiment -- post-commit per-task failure analysis for a specific experiment; reads per-task traces and the eval-runner log, writes per-task…
ideator
Generates ranked experiment proposals for the evo orchestrator. Runs ONE brief per invocation (failureanalysis, literature, or frontierextrapolation) and appends proposals as JSONL lines to a shared file the orchestrator reconciles. Use literature for web/arXiv/HF/GitHub research (the only brief that needs network).…
verifier
Read-only audit of one evo experiment for design-time cheating (pre-phase) or result-time validity (post-phase). Catches test-set leakage in training data, subsetted eval commands, missing gates for new artifacts, generic hypotheses, cache short-circuits, fake artifacts, and score-reproducibility failures. Returns…
hypothesis
Analysiere Eval-Failures / Metrik-Ergebnisse und generiere eine testbare Verbesserungshypothese.
mutator
Wende eine Hypothese als gezielte Änderung auf die Zieldateien an.
orchestrator
Koordiniert den Agent-Lifecycle im Skill Forge Loop.