reflect

A command that manually starts the agent-harness self-review process, with an optional dry run that shows proposed changes without applying them.

In plain words
What is it for?
Use it to start a meta-reflection immediately or from an external script, and to preview the proposed improvement safely.
Why use it?
It lets you review or schedule the self-improvement process without waiting for its normal five-run trigger, while dry-run mode avoids changing files or creating commits.

Command

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 commands/stilero/claude-plugins/reflect
Clone the repo
git clone --depth 1 https://github.com/stilero/claude-plugins
Per session 22 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 530 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 $0.00022 $0.00530
Opus 5 $0.00011 $0.00265
Sonnet 5 $0.00004 $0.00106
Haiku 4.5 $0.00002 $0.00053

Measured yesterday against content hash d5f1fca917c6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

reflect 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 yesterday.

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/commands/reflect.md · 32 lines

How it starts

The opening of the file, as written. The whole thing — 32 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are invoking the meta-reflector manually. The agent-harness plugin's self-improvement loop is normally count-triggered (every 5 completed runs from the orchestrator's Stage 5), but this command lets you force a meta-run on-demand or scheduled externally via scripts/run-meta.sh.

Process

  1. Inspect the slash-command argument string. If $ARGUMENTS contains the token --dry-run:

    • Dispatch the meta-reflector subagent (via the Task tool) with this instruction prepended to its normal context:

      DRY RUN: produce the reflection file and show the intended change as a diff, but do NOT edit any agent or config file, do NOT commit anything to the plugin repo, and do NOT open any PR. Print the proposed diff in the reflection's ## Change applied section as a fenced diff block. Skip Step F (stats.md rewrite) so runs_since_last_meta stays at its current value.

    • Otherwise (no --dry-run): dispatch the meta-reflector subagent with no extra prefix. It will follow its own algorithm (Steps A–G in agents/meta-reflector.md).

  2. When the subagent returns, parse its JSON output ({ "reflection_path": "...", "change_applied": true|false, "verdict_prior": "..." }).

  3. Report to the user, in two short paragraphs:

    • The reflection file path.
    • What changed if anything (commit SHA + summary), or "no change applied" with the reason.
    • The verdict on the prior change (SUSTAIN, FLAG, AUTO-REVERT, AUTO-REVERT-FAILED, or no-prior-change).
    • Any workflow-proposal PRs opened, with URLs.

Hard rules

  • You do NOT execute the meta algorithm yourself. You only dispatch the subagent and relay its output. The subagent is the one with the safety constraints (one-change-per-run, AUTO-REVERT gate, oscillation check, lint-before-commit, max_model ceiling, draft-only PRs).
  • Do NOT edit any file from this command. The subagent writes its own reflection and applies its own change.
  • If the subagent returns malformed output (no JSON block, missing fields), report the failure verbatim and do not attempt recovery — the next meta-run will see the unchanged state and try again.

Read the full file on GitHub · 32 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. yesterday First seen · 32 lines · 0 tokens per session scan A d5f1fca917c6

Subscribe to this mod's changes

reflect is a command published in the GitHub repository stilero/claude-plugins (2 stars, last pushed 2mo ago), licensed MIT. It adds 22 tokens to every session and 530 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.