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.
git clone --depth 1 https://github.com/topprismdata/cultivating-ml-agentWrote 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/commands/topprismdata/cultivating-ml-agent/meta-apply)<a href="https://agentmods.dev/commands/topprismdata/cultivating-ml-agent/meta-apply"><img src="https://agentmods.dev/badge/commands/topprismdata/cultivating-ml-agent/meta-apply.svg" alt="Measured on agentmods" 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 | $0.00019 | $0.00580 |
| Opus 5 | $0.00010 | $0.00290 |
| Sonnet 5 | $0.00004 | $0.00116 |
| Haiku 4.5 | $0.00002 | $0.00058 |
Grade A, and why
meta-apply 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 8d 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 — 86 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/meta-apply — Apply Memory Updates
Gated memory writes. Sends proposed changes to cross-model jury (agy/gemini/codex/ollama) for approval. Only lands changes the jury validates.
Usage
/meta-apply # Interactive: review each suggestion
/meta-apply --all # Apply all after jury approval
/meta-apply <file> # Apply to specific file
The Flow
1. /meta-optimize runs analysis
2. You review the report
3. /meta-apply proposes changes
4. Each change is sent to cross-model jury (agy)
5. Jury says: APPROVE / REJECT / REVISE
6. Only APPROVED changes land
7. Failed changes get logged to memory/meta-apply-rejected.md
Why Jury Gating
From ARIS design:
"Self-evolution layer (
/meta-optimize): analyzes logs and proposes SKILL.md patches; now read-only with landing gated by cross-model jury via the new/meta-applyskill"
Without a jury:
- Agent reinforces its own errors
- Stale principles never get challenged
- Conflicts slip through
When to Use
- After
/meta-optimizereports issues you agree with - When you want to update skills based on new evidence
- When refreshing dead-end entries
When NOT to Use
- During an active competition (memory changes are noise)
- On principles you just wrote (let them age first)
- If the changes are trivial formatting (overhead)
Jury Output Format
The jury receives:
PROPOSED CHANGE: <file>
<before>
<after>
REASON: <why this change>
And returns one of:
- APPROVE — apply as-is
- REJECT — don't apply, here's why
- REVISE — here's a better version
Provenance Tracking
Every approved change is logged to memory/meta-apply-log.md:
## 2026-06-01
- File: memory/skills/500-line-rule.md
- Change: added example
- Author: meta-optimize
- Jury: agy (gemini-1.5-pro)
- Status: APPROVED
This ensures auditability: "who changed what when".
Anti-Patterns
- ❌
--allwithout reading the report (rubber-stamping) - ❌ Ignoring REJECT verdicts (defeats purpose)
- ❌ Skipping provenance tracking
- ❌ Updating principles you just wrote (insufficient reflection time)
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.
- 8d ago First seen · 86 lines · 19 tokens per session scan A c6362d7ce2e5
meta-apply is a command published in the GitHub repository topprismdata/cultivating-ml-agent (5 stars, last pushed 10d ago), licensed MIT. It adds 19 tokens to every session and 580 once invoked, about $0.0001 per session on Opus 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 2026-08-31.
Other commands, from other repositories
tree-ring-recall
Recall durable Tree Ring Memory context before starting or resuming work.
tree-ring-update
Check for or install a verified Tree Ring Memory CLI update without changing installation scope.
tree-ring-certify
Generate Tree Ring harness or recall-quality evidence without confusing it with the full framework release suite.
tree-ring-status
Check receipt-backed Tree Ring harness readiness without claiming configuration is activation.
unforgit-curate
Review and improve Unforgit memory quality.
remember
Save text as a project memory in Rembric.