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 inclusionAI/Avernet --skill clawevolve-reviewgit clone --depth 1 https://github.com/inclusionAI/AvernetWrote 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/inclusionai/avernet/clawevolve-review)<a href="https://agentmods.dev/skills/inclusionai/avernet/clawevolve-review"><img src="https://agentmods.dev/badge/skills/inclusionai/avernet/clawevolve-review/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/inclusionai/avernet/clawevolve-review"><img src="https://agentmods.dev/badge/skills/inclusionai/avernet/clawevolve-review.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 | $0.00068 | $0.01249 |
| Opus 5.5 | $0.00027 | $0.00500 |
| Sonnet 5.5 | $0.00014 | $0.00250 |
| Haiku 4.5 | $0.00007 | $0.00125 |
Grade A, and why
clawevolve-review 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 — 128 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ClawEvolve Structured Review v2
职责边界
Review 只回答:
- 需要区分什么 failure mechanism?
- 有哪些仍然 plausible 的替代原因?
- 什么 capability-level 行为信号能区分这些解释?
- 哪些 accepted baseline 行为必须保护?
- 当前 hypothesis 是 suspected、testing、supported 还是 falsified?
- 什么新证据出现后应重新考虑?
Review 不回答:
- 具体怎么改;
- 使用哪个 mutation operator;
- 修改哪个文件;
- exact patch / proposed change;
- 增加、删除或重排什么 Block/anchor/步骤文案。
具体实现决策属于 Tune。
可读取输入
- objective.md;
- 输入 evolution.spec.v1 的 experiment questions、protected behaviors 和 contracts;
- post-Tune full candidate optimization report;
- candidate effect/protected/full-opt gate;
- Tune report、change manifest 与 diff;
- aggregate validation summary;
- acceptance / replication decision;
- runtime experiment ledger 和 failure registry 中的机制级信息。
Validation 防火墙
- validation 只用于 promotion/generalization 门禁,不是下一轮训练数据。
- 禁止读取 validation resultPath、逐 task report、transcript、case/template、rubric breakdown。
- 禁止输出 validation task id、单 case 分数、expert/rubric 字段、答案、原文或 case-specific workaround。
- 只能使用 runner 注入的 aggregate validation 与抽象 acceptance decision。
效果解释原则
accepted/rejected描述的是 promotion 结果,不等同于effective/ineffective。即使未晋升,也要分别记录已经观察到的正向行为、回归行为和证据等级。- 不得只根据 aggregate score 判断机制。分别审视:
reasoning_quality:识别、准确度、证据和建议;delivery_reliability:最终输出完整性、确认、stage completion、非重复和一次性交付。
- 中段推理提升但尾段交付消失,应描述为
reasoning-only improvement with delivery regression,不能写成整体成功。 - 高分但存在重复区块、尾段未完成或 judge/breakdown 缺失时,必须保留 scorer blind spot / observability alternative cause。
- 单次 targeted observation 默认只记为
observed_once;只有重复、跨任务或 held-out aggregate 证据才逐步升级为replicated、cross_task_positive、heldout_positive。 - Review 应把已观察到的成功机制转成 capability-level protected behavior,避免后续 Tune 在解决新问题时无意撤销;但不得把候选的具体 patch 文案写入 spec。
唯一输出
只写:
{round_dir}/spec/review_decision.json
不得直接写 spec-vN.md、spec-vN.json 或 spec_update_report.md;这些由 runner 确定性生成。
Schema:
{
"schema_version": "evolution.review_decision.v2",
"round_id": 1,
"acceptance_decision": "needs_replication",
"summary": "机制级短总结",
"confidence": "low",
"hypotheses": [
{
"hypothesis_id": "HYP-001",
"failure_signature": "complex_case_stage_abandonment",
"status": "suspected",
"claim": "复杂任务中关键路径过长可能导致后续阶段放弃",
"alternative_causes": ["instruction_density", "output_budget", "tool_latency"],
"disambiguation_signal": "保持业务目标不变时,stage_completion 是否稳定恢复",
"protected_behaviors": [{"behavior_id": "complete_output_contract", "description": "保持完整输出契约", "metric": "output_contract_complete", "direction": "maintain", "min_value": 1, "max_drop": 0}],
"confidence": "low",
"revisit_condition": "独立重复实验产生一致证据"
}
],
"direction_decisions": [
{
"direction_id": "DIR-001",
"decision": "keep|strengthen|split|freeze|reject|defer",
"confidence": "low",
"rationale": "机制级理由",
"revisit_condition": "重新考虑条件"
}
]
}
What ships with it
5 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 · 128 lines · 68 tokens per session scan A 3abd39520a6b
clawevolve-review is a skill published in the GitHub repository inclusionAI/Avernet (641 stars, last pushed today), licensed Apache-2.0. It adds 68 tokens to every session and 1,249 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 2026-09-15.
Other skills, from other repositories
investigate-sentry-issue
Investigate and triage a Sentry error issue.
sre-agent
Deploy or update the SRE Agent on Omnara, a production investigator that diagnoses issues across AWS and source code and reports the likely cause with evidence. Use when the user asks to deploy, set up, update, or remove the SRE agent.
apps
Build and iterate on schema-backed Swarm Apps with live models, named queries, custom actions, and json-render interfaces using the app-upsert, app-get, app-list, and app-patch tools.
kapso-whatsapp
How to interact with Kapso WhatsApp from the swarm — read inbound webhook payloads (text AND media), fetch message history, send free-form messages within the 24h session window (and template messages outside it), mark-as-read, show the typing indicator, send reactions, download media, verify webhook signatures, and…
swarm-scripts
Bulk, repeat, fan-out, or data-heavy work: write and run swarm scripts (inline script-run, named script-upsert, durable launch-script-run). Covers the script-vs-tool rubric, the authoring contract (args first, ctx second), the seed catalog, connections and secrets, dbquery, and exposing a script as an API.
composio
Use Composio from Agent Swarm through the agent-swarm x composio CLI route, the swarmx MCP tool, or a registered ctx.api.composio script connection. Trigger when a task needs connected third-party app tools such as Gmail, Google Calendar, Google Docs, Google Drive, GitHub, Slack, Notion, or HubSpot through Tool Router…