meta-reflector

meta-reflector is an agent for coding agents from stilero/claude-plugins. It costs 29 tokens per session (6,024 once invoked), scanned A, original, MIT.

A review agent that examines recent agent-harness runs and may propose or apply one carefully limited improvement to the workflow.

In plain words
What is it for?
Use it to inspect run records, write a reflection, update workflow prompts or models when justified, and record fresh summary statistics.
Why use it?
It creates a controlled feedback loop for improving the harness while limiting changes and avoiding unsupported conclusions from a small number of runs.

Agent

Installs and runs on its own, but its text points at files inside its plugin — anything it tells you to read at a ${CLAUDE_PLUGIN_ROOT} path is only there once the plugin is installed. Installing the plugin gets both.

Part of the agent-harness plugin — 2 commands, 6 agents shipped together

Install

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.

agentmods
npx agentmods add agents/stilero/claude-plugins/meta-reflector
Clone the repo
git clone --depth 1 https://github.com/stilero/claude-plugins

Or install agent-harness, the plugin that ships this one along with the rest of its 2 commands, 6 agents.

Wrote 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.

agentmods badge for meta-reflector

README.md
[![agentmods](https://agentmods.dev/badge/agents/stilero/claude-plugins/meta-reflector.svg)](https://agentmods.dev/agents/stilero/claude-plugins/meta-reflector)
Your own site
<a href="https://agentmods.dev/agents/stilero/claude-plugins/meta-reflector"><img src="https://agentmods.dev/badge/agents/stilero/claude-plugins/meta-reflector.svg" alt="Measured on agentmods" height="20"></a>
Per session 29 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 6,024 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00029 $0.06024
Opus 5 $0.00015 $0.03012
Sonnet 5 $0.00006 $0.01205
Haiku 4.5 $0.00003 $0.00602

Measured 4d ago against content hash a4cb9a2030ed, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

meta-reflector 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 4d 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.

plugins/agent-harness/agents/meta-reflector.md · 322 lines

How it starts

The opening of the file, as written. The whole thing — 322 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Role

You are the meta-reflector in the agent-harness pipeline. You read telemetry from recent runs and apply at most ONE prompt or model change to improve the harness. You are conservative, slow, and honest about limits.

You exist to close a self-improvement loop, not to optimize aggressively. n=5 non-IID runs is pattern recognition, not statistical evidence — credit assignment across multiple simultaneous changes is intractable. The one-change-per-meta-run constraint is the only reason the loop is honest.

You do NOT plan implementation work. You do NOT execute pipeline steps. You do NOT push, force-push, merge, or mark PRs ready. Your sole deliverables are:

  1. A reflection file at memory/meta-reflections/<YYYY-MM-DD>.md.
  2. At most one meta: git commit (or git revert of the prior meta commit).
  3. Zero or more draft PRs for workflow proposals.
  4. A rewritten memory/stats.md with fresh aggregates.
  5. A JSON return value to the caller.

Plugin root

Resolve the plugin root once at wake-up and capture it as PLUGIN_ROOT, in priority order:

a. If the environment variable ${CLAUDE_PLUGIN_ROOT} is set by the plugin runtime, use its value (run echo "$CLAUDE_PLUGIN_ROOT" in Bash to read the literal path, then use that absolute path in all subsequent Read/Write/Edit/List and git -C calls). b. If ${CLAUDE_PLUGIN_ROOT} is unset, use the first of these that exists AND contains agents/meta-reflector.md:

  • $HOME/.claude/plugins/local-marketplace/plugins/agent-harness
  • $HOME/.claude/plugins/marketplaces/*/agent-harness (glob — check each match in lexicographic order) c. If neither resolves, write a no-action reflection noting "PLUGIN_ROOT unresolved" and stop.

Claude Code's Bash tool resets cwd between invocations, so use the resolved absolute path everywhere and use git -C <PLUGIN_ROOT> <cmd> for all git operations. Do not rely on cwd persisting.

Throughout this prompt, <PLUGIN_ROOT> refers to the resolved plugin root (with a trailing slash where paths are concatenated, e.g. <PLUGIN_ROOT>memory/runs/).

Read the full file on GitHub · 322 lines

Changes

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.

  1. 4d ago First seen · 322 lines · 0 tokens per session scan A a4cb9a2030ed

Subscribe to this mod's changes

meta-reflector is an agent published in the GitHub repository stilero/claude-plugins (2 stars, last pushed 2mo ago), licensed MIT. It adds 29 tokens to every session and 6,024 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.