Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/try-works/recursive-modenpx agentmods add skills/try-works/recursive-mode/recursive-trainingWrote 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/try-works/recursive-mode/recursive-training)<a href="https://agentmods.dev/skills/try-works/recursive-mode/recursive-training"><img src="https://agentmods.dev/badge/skills/try-works/recursive-mode/recursive-training/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/try-works/recursive-mode/recursive-training"><img src="https://agentmods.dev/badge/skills/try-works/recursive-mode/recursive-training.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00034 | $0.01636 |
| Opus 5.5 | $0.00014 | $0.00654 |
| Sonnet 5.5 | $0.00007 | $0.00327 |
| Haiku 4.5 | $0.00003 | $0.00164 |
Grade A, and why
recursive-training 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 — 196 lines — stays where its author put it; the contents beside it link to each section on GitHub.
recursive-training
Purpose
Use this skill after a repository has accumulated completed recursive-mode runs and you want to turn repeated successes or failures into durable, repo-local guidance.
The canonical workflow still lives in /.recursive/RECURSIVE.md. This skill owns only the training, loading, and memory-discipline layer that sits around completed runs.
When to use it
Use recursive-training to:
- extract cross-run learnings from completed recursive-mode runs
- keep those learnings in
/.recursive/memory/instead of ad hoc mirrors - refresh memory after Phase 8 locks
- load only the most relevant prior learnings before a new run starts
- provide startup guidance without mutating the memory plane
Hard rules
- Repository-local only. Training data and extracted memory stay inside the current repo.
- No parameter updates. Learning happens through files in
/.recursive/memory/, not model mutation. /.recursive/memory/is the only canonical store. Pointer files are bootstrap-managed and non-authoritative.- All markdown under
/.recursive/run/<run-id>/is eligible training input, not just00-08. - Group runs by subsystem only. The extractor assigns task types per learning item.
- Use contrastive extraction when both winners and losers exist; fall back to winner-only extraction for high-quality repos.
- Every extracted item must remain evidence-grounded in completed runs.
- Training scripts do not own
AGENTS.md; bootstrap owns bridge-file updates.
Training model
recursive-training combines two ideas:
- ReasoningBank-style memory items for structured, reusable extracted guidance
- Training-free GRPO-style comparison for contrastive winner/loser extraction when variance exists
At a high level:
- Parse all markdown artifacts from completed runs.
- Infer the dominant subsystem from changed paths and evidence across those artifacts.
- Classify each subsystem group as contrastive, winner-only, or insufficient.
- Ask the extractor for structured learning items.
- Write those items into:
/.recursive/memory/domains/<subsystem>.md/.recursive/memory/training/<task-type>.md
- Refresh the memory registry/startup guidance without treating pointer files as authoritative memory.
What ships with it
3 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 · 196 lines · 34 tokens per session scan A f6a9016c7da4
recursive-training is a skill published in the GitHub repository try-works/recursive-mode (136 stars, last pushed 3mo ago), licensed Apache-2.0. It adds 34 tokens to every session and 1,636 once invoked, about $0.0001 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 2026-08-30.
Other skills, from other repositories
knowledge-ingest
Ingest documents, articles, conversations, or supplied text into Knowledge with retained originals, sourced claims, entity updates, timelines, and cross-links; also learn durable lessons from a set of conversations, transcripts, or past sessions (/learn), including the Learn from conversations and Learn from Memory…
mem0-oss-to-platform
Plan, then execute, a migration of a project from the mem0 open-source / self-hosted SDK (local Memory class) to the mem0 Platform / hosted SDK (MemoryClient). TRIGGER when: the developer wants to move, switch, or migrate mem0 off OSS/self-hosted to the hosted API, e.g. "migrate my mem0 setup to the platform", "switch…
forget
Remove a personal fact, preference, relationship detail, topic, or prior event from Muse's active memory and stop existing copies or automations from bringing it back. Use for explicit requests such as 'forget that', 'don't remember this about me', or 'remove that from your memory'. Do not use when 'forget it' merely…
Cortex
LifeOS memory, prior-session recall, and learning capture. USE WHEN capture this learning, document this learning, what did we learn recently, recent learnings, cortex, knowledge, knowledge base, search knowledge, what do we know about, archive, harvest, knowledge status, develop note, add to knowledge, add a people…
memory
Use when the user asks to remember, recall, forget, update, search, or inspect durable OpenSquilla memory, including profile facts in USER.md and long-term notes in MEMORY.md or memory//.md.
ha-data-stores
Map of Hope Agent's local data stores and safe read-only query workflow. Use when the user asks where Hope Agent stores data, wants to inspect sessions/messages/memory/logs/background jobs/knowledge indexes/settings, asks the model to query local app data, or debugging requires checking persisted state. Trigger…