try-works/role-model / .agents/skills/recursive-mode/skills/recursive-training
recursive-training
VerifiedUse after completed recursive-mode runs accumulate to extract durable experiential memory into `/.recursive/memory/`, then load it for later runs through the canonical loader.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.7 KB |
| agents/openai.yaml | config | 270 B |
| references/memory-architecture.md | markdown | 3.6 KB |
| references/phase8-and-loading.md | markdown | 3.1 KB |
references/memory-architecture.md
View on GitHub# recursive-training memory architecture
## Architecture overview
```text
.recursive/run/<run-id>/ -> parse -> rollout summaries
any *.md file |
v
group by subsystem
|
v
classify group: contrastive | winner-only | insufficient
|
v
extractor emits structured learning items
|
v
/.recursive/memory/domains/<subsystem>.md
/.recursive/memory/training/<task-type>.md
```
The extractor writes every learning item to the domain shard for its subsystem, then fans those same items out to task-type training shards.
## Structured memory schema
Each extracted item should include:
- `title`
- `description`
- `content`
- `task_type`
- `subsystem`
- `source_runs`
- `applies_to`
- `created_at`
- `success_rate`
The intent is a compact, reusable memory item rather than a long narrative dump.
## Training data source
The parser treats **all** markdown in each completed run folder as eligible evidence, including:
- `00-requirements.md`
- `01-as-is.md`
- `02-to-be-plan.md`
- `03-implementation-summary.md`
- `04-test-summary.md`
- `05-manual-qa.md`
- `06-decisions-update.md`
- `07-state-update.md`
- `08-memory-impact.md`
- audit docs, addenda, evidence summaries, and any other agent-authored markdown
## Evidence signals
Reward and classification signals come from the artifacts themselves, including:
- audit verdicts
- coverage and approval gates
- QA verdict or equivalent approval markers
- build/typecheck pass or fail
- test counts and pass/fail signals
- evidence of implementation and verification across the run artifacts
If a run has neither implementation evidence nor test evidence anywhere in its markdown set, it should be treated as critically incomplete.
## Grouping strategy
Group runs by **subsystem only**.
Why:
- one run often teaches multiple lessons at once
- `00-requirements.md`, `02-to-be-plan.md`, `03-implementation-summary.md`, and `04-test-summary.md` can all yield different task-type learnings
- grouping by task type too early would throw away those cross-cutting signals
The extractor, not the grouping stage, decides each learning item's `task_type`.
## Extraction modes
| Mode | When to use it | Goal |
| --- | --- | --- |
| Contrastive | At least one winner and one loser in the same subsystem group | Explain why successful runs worked where failed runs did not |
| Winner-only | Enough successful runs but no losers | Capture repeated successful conventions and repo quirks |
| Insufficient | Too little evidence or too little variance | Skip extraction |
## Output contract
After a successful extraction pass:
- subsystem items live in `/.recursive/memory/domains/<subsystem>.md`
- task-type items live in `/.recursive/memory/training/<task-type>.md`
- `/.recursive/memory/MEMORY.md` remains the discovery router
- pointer files such as `.cursorrules`, `CLAUDE.md`, and `.github/copilot-instructions.md` stay non-authoritative
Files never modified by training:
- assistant internals
- files outside the current git working tree
- locked earlier-phase artifacts
- `AGENTS.md` bridges
## Script boundary
Training scripts own prompt construction, extraction orchestration, parsing, and memory-file writes.
They do **not** own:
- bridge-file updates
- transport-selection policy
- overall recursive-mode phase orchestration
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |