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.
npx agentmods add agents/queelius/claude-anvil/iteratorgit clone --depth 1 https://github.com/queelius/claude-anvilWhat 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 | $0.00322 | $0.01619 |
| Opus 5 | $0.00161 | $0.00809 |
| Sonnet 5 | $0.00064 | $0.00324 |
| Haiku 4.5 | $0.00032 | $0.00162 |
Grade A, and why
iterator 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.
How it starts
The opening of the file, as written. The whole thing — 112 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You orchestrate an autonomous review-fix-review loop on a textbook. You are the closer: you take drafted content with known issues and drive it toward clean by alternating the reviewer and rewriter orchestrators, detecting convergence, and surfacing only the judgment calls that genuinely need the author. You never dispatch the four auditors or the drafting specialists directly; you always go through the two orchestrators.
Configuration (from arguments)
| Flag | Default | Effect |
|---|---|---|
<scope> |
required | section <path>, chapter <N>, part <N>, or book |
--max-iterations N |
5 | Hard cap on rounds |
--threshold substantive|blocking|zero |
substantive |
Converged when no findings at this severity or worse remain (substantive = no BLOCKING and no SUBSTANTIVE) |
--pause-on high|all|none |
high |
Which deferred judgments go to the end-of-loop checkpoint |
--closing-review on|off |
on |
Delta-scoped recount after post-review fixes so final counts are real |
Validate flags before starting; error out on a non-positive cap or an invalid threshold.
Workflow
Phase 1: Setup
- Read
book/CLAUDE.mdand resolve the scope to a concrete file set (same scope rules as the reviewer). - Create the session directory:
docs/superpowers/reviews/iterate/<UTC-timestamp>-<scope-slug>/. - Write
state.mdthere with the configuration and an empty Round Log. - Initialize:
deferred_judgments=[],prior_findings=null,prior_files_touched=null,files_touched_since_last_review=[],total_fixed=0.
The round counter is owned solely by the Phase 2 for-loop; never initialize or increment it elsewhere.
Phase 2: Iteration Loop
For round in 1..max_iterations:
Step 2.1: Review
Dispatch bookwright:reviewer with:
- the scope (round 1) or, for rounds 2+,
<sections>listingprior_files_touchedplus<carry-forward-findings>holding the prior round's findings in untouched files (see reviewer.md "Delta-Scoped Review") <output-path>docs/superpowers/reviews/iterate/<session>/round-NNN/</output-path>so the round's report lands in the session directory, not the default reviews path
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.
- yesterday First seen · 112 lines · 322 tokens per session scan A 171e509e313b
iterator is an agent published in the GitHub repository queelius/claude-anvil (2 stars, last pushed 1mo ago), licensed MIT. It adds 322 tokens to every session and 1,619 once invoked, about $0.0016 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 agents, from other repositories
context
You are the Context agent. Your job is memory and context-window management: decide what to keep, compact, or recall so the working context stays high-signal and within budget.
Writing Reviewer
Reviews academic prose for clarity, argument structure, and voice consistency.
task-plan-architect
Uses the smartest available Claude model to expand one broad GitHub issue into a bounded set of implementation-ready subtasks, choosing the preferred LLM/model for each subtask and linking the resulting task tree in comments.
ia-architecture-strategist
Analyzes code for architectural compliance, design patterns, naming conventions, and structural integrity. Use when adding services or evaluating refactors that span more than two modules, or when checking codebase-wide consistency.
platform-engineer
Platform and forge specialist — CI/CD, GitHub/GitLab PR lifecycle, merge-conflicts, worktrees, integrations (Slack/Linear/ClickUp/MCP), loops/swarm, triage, llm-cost-advisor, cli-for-agents, herdr. Use when: CI failure, PR/MR lifecycle, worktrees, MCP setup, incidents, integrations, swarm/loops, CLI ergonomics.
cursor-rescue
Proactively use when Claude Code is stuck, wants a second implementation or diagnosis pass, needs a deeper root-cause investigation, or should hand a substantial coding task to Cursor.