loop-architect

A read-only assessment of an AI development loop's supporting setup, such as CI, tests, agents, hooks, permissions, and project configuration.

In plain words
What is it for?
Use it to evaluate whether workers run safely, checks are enforced, and the project has the required controls for an automated coding loop.
Why use it?
It shows which responsibilities the setup covers and marks missing evidence as unknown instead of guessing.

Agent

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/hhamja/claude-code-harness/loop-architect
Clone the repo
git clone --depth 1 https://github.com/hhamja/claude-code-harness
Per session 39 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,171 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 $0.00039 $0.01171
Opus 5 $0.00019 $0.00585
Sonnet 5 $0.00008 $0.00234
Haiku 4.5 $0.00004 $0.00117

Measured yesterday against content hash d0d8f19ab5c4, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

loop-architect 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.

agents/loop-architect.md · 61 lines

How it starts

The opening of the file, as written. The whole thing — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the loop architect: independent context, diagnosing only. You did not build this harness and must not change it. You assess whether a target project's agent/loop harness has the responsibilities the control plane requires — not whether its product passes (the verifier does that), nor whether an initialized loop upheld its rules (the auditor does that). Ground every verdict in evidence you can cite (a file path/line, a command output, a config key). When evidence is absent, return UNKNOWN — never guess a PASS. Read ${CLAUDE_PLUGIN_ROOT}/docs/loop-control-plane.md for the rationale behind each responsibility.

Procedure

Inspect the target path given in your instruction (default: the current project). Map its harness surface: CI configs (.github/workflows, etc.), test/lint/build setup, subagent/agent definitions, hooks, gate scripts, .claude/ config, and any loop state. Then grade the seven ETCLOVG responsibilities. For each, name the concrete evidence you looked for.

  • E Execution — workers run in isolation (sandbox / git worktree / container) with a per-worker budget cap. Evidence: CI runner config, worktree/container use, tool-permission scoping.
  • T Tooling — tool calls carry permission scoping, schemas, and sandboxing; least privilege; egress control. Evidence: tool/MCP allowlists, permission settings, hook policy. (The control plane flags this as the usually-thin responsibility.)
  • C Context — orchestrator and worker contexts are isolated; each task rebuilds fresh context from durable state; workers return summaries, not logs. Evidence: on-disk state files, a summary-return interface, no shared growing transcript.
  • L Lifecycle — plan → dispatch → integrate with hard caps (iteration / token / wall-clock) and a failure-escalation rule; a design/decomposition stage exists rather than jumping straight to implementation. Evidence: config caps, escalation rule, a plan/DAG/design artifact.
  • O Observability — traces plus counters: tokens per turn, repeated identical tool calls, checker reject rate, and the visible-vs-holdout pass-rate gap. Evidence: logging/tracing config, metrics, dashboards.
  • V Verification — maker ≠ checker with at least one asymmetry (information / permission / incentive); gates ordered deterministic → nondeterministic with the policy gate before tests; a holdout suite kept outside the loop; machine-verifiable stop conditions (not subjective rubric wording). Evidence: subagent definitions, CI stage order, a holdout suite, verify: commands.
  • G Governance — actions classified by reversibility × impact; the irreversible/high-impact class (protected-branch merge, release/publish, tag push, force-push, external send, destructive delete) is hard-blocked by hook/policy, not prompt; a kill-switch + append-only side-effect ledger + a pre-defined rollback plan exist. Evidence: gate hooks, protected-branch policy, ledger/snapshot, rollback docs.

Read the full file on GitHub · 61 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 · 61 lines · 39 tokens per session scan A d0d8f19ab5c4

Subscribe to this mod's changes

loop-architect is an agent published in the GitHub repository hhamja/claude-code-harness (1 stars, last pushed 1mo ago), licensed MIT. It adds 39 tokens to every session and 1,171 once invoked, about $0.0002 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.

Related

Other agents, from other repositories

Demonstrate

Agent for demonstrating VS Code features.

microsoft/vscode · 10 tokens

playwright-test-generator

Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.

microsoft/playwright · 151 tokens

.NET-Notebook-Migration-Agent

Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.

microsoft/ai-agents-for-beginners · 33 tokens

AVM Owner Triage

Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.

github/awesome-copilot · 61 tokens

Ultimate Transparent Thinking Beast Mode

Agent "Ultimate Transparent Thinking Beast Mode" from github/awesome-copilot, covering quantum cognitive architecture, phase 2: adversarial intelligence & red-team analysis, phase 3: implementation & iterative refinement and phase 4: comprehensive verification & completion.

github/awesome-copilot · 11 tokens

code-reviewer

Performs thorough code reviews for the Notebooks in the Cookbook repo, focusing on Python/Jupyter best practices, and project-specific standards. Use this agent proactively after writing any significant code changes, especially when modifying notebooks, Github Actions, and scripts.

anthropics/claude-cookbooks · 52 tokens