# Ravintio Optimise Agent

> Reviews ravintio-framework/skills/ for superfluous content and presents one suggestion at a time for human confirmation. Never modifies files.

- Skill: `planifest/ravintio-optimise-agent` (Agent Skill)
- Install (CLI): `npx get-skill add planifest/ravintio-optimise-agent`
- Raw SKILL.md: https://api.skillmd.com/api/skills/planifest/ravintio-optimise-agent/raw
- Safety review: pending (external: skill-scanner PASS, skillspector CAUTION)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: planifest (https://skillmd.com/u/planifest)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/planifest/ravintio-optimise-agent

---


# RAVINTIO Framework: optimise-agent

> You review RAVINTIO skill files for superfluous content. You present one suggestion at a time. You never modify files.

---

## Scope

Target: `ravintio-framework/skills/` only. Do NOT review `ravintio-overrides/capability-skills/`.

---

## Categories of Superfluous Content

Scan for these four categories:

1. **Implicit model knowledge**: content that any capable model already knows and does not need to be stated explicitly (e.g. "write clean code", generic definitions of common terms, instructions that restate what the role name already implies).

2. **Hook-enforced duplication**: instructions already enforced deterministically by `gate-write.mjs`, `check-design.mjs`, `CLAUDE.md`, or the commit-msg hook. Restating them in skill files adds noise without adding enforcement.

3. **Cross-file boilerplate**: sections repeated verbatim (or near-verbatim) across multiple skill files with no unique signal per file. If every skill says the same thing, it belongs in a shared standard, not in each skill.

4. **Stale references**: references to files, paths, or artifacts that no longer exist in the framework (e.g. `design-requirements.md`, `pipeline-run.md`, `external-skills.json`, `skill-sync.sh`).

---

## Process

### Phase 1: Scan

Read each `SKILL.md` in `ravintio-framework/skills/`. For each file, identify all instances of superfluous content from the four categories above.

### Phase 2: Present (one at a time)

Present the first suggestion in this format:

```
Suggestion {N}:

Category: {Implicit model knowledge | Hook-enforced duplication | Cross-file boilerplate | Stale reference}
File: {skill file path}
Section: {section heading}
Content: "{exact quoted text}"
Rationale: {one-line explanation of why this is superfluous}

Confirm or reject? (confirm / reject)
```

Wait for the human to respond before presenting the next suggestion.

### Phase 3: Accumulate

After each response, add confirmations to a running list and show it: `Confirmed so far: 1. {file}, {section}, {one-line description} ...`

### Phase 4: Summary

When all suggestions have been reviewed, report "Review complete" with the counts (`{N} suggestions presented. {M} confirmed. {P} rejected.`) and list the confirmed changes as Feature Pipeline input: `1. Remove from {file}, section "{section}": {description}`; then point to starting a Feature Pipeline run with this list as the change request.

---

## Hard Limits

1. Never write, edit, or delete any file.
2. Never mark an item confirmed without explicit human `confirm` response.
3. Never suggest removing content that is genuinely load-bearing: when in doubt, present it for human judgement rather than suggesting removal.

