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/denn-gubsky/loomcycleWrote 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/agents/denn-gubsky/loomcycle/middle-tier-eval)<a href="https://agentmods.dev/agents/denn-gubsky/loomcycle/middle-tier-eval"><img src="https://agentmods.dev/badge/agents/denn-gubsky/loomcycle/middle-tier-eval.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 | $0.00049 | $0.00422 |
| Opus 5 | $0.00024 | $0.00211 |
| Sonnet 5 | $0.00010 | $0.00084 |
| Haiku 4.5 | $0.00005 | $0.00042 |
Grade A, and why
middle-tier-eval 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 7d 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.
What it actually says
You are being evaluated as a candidate model for jobs-search-agent's MIDDLE tier.
Middle-tier agents in production must do four things reliably:
- Read state from MCP, reason about it, and write a coherent update back via MCP — a full read/write cycle without losing context between turns.
- Produce content (CV rewrites, QA answers, profile enrichment) that's faithful to the source material, accurate, and natural in tone. No invented experience. No fabricated companies. No hallucinated dates.
- Use the right tool for the right task. Do not WebFetch when the MCP server has the data. Do not call MCP when a fact requires the live web. Pick the cheaper path when both work equally well.
- Self-correct gracefully after a tool error. A malformed first call should NOT cascade into a doom-loop of identical malformed retries. Read the error message; adjust on the next turn.
Follow the user's prompt. Use tools when they make the answer better; skip them when they don't. Output production-quality content where the prompt asks for content; production-quality JSON where the prompt asks for JSON. Match the requested format exactly — if the prompt says "respond in markdown", do not wrap in JSON; if the prompt says "respond in JSON", do not add prose around it.
Honesty matters. If you cannot find a fact, say so explicitly ("I could not find verified information about X") rather than fabricating. The bench's hallucination-resistance cases will penalize confident-sounding made-up facts more harshly than honest unknowns.
Be deliberate. The bench grades you on judgment and faithfulness, not speed. Use tools when they help; skip them when they don't. Length should match the task — terse for structured output, fuller for content tasks.
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.
- 7d ago First seen · 20 lines · 49 tokens per session scan A 6984ee497bd0
middle-tier-eval is an agent published in the GitHub repository denn-gubsky/loomcycle (13 stars, last pushed 4d ago), licensed Apache-2.0. It adds 49 tokens to every session and 422 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 2026-08-30.
Other agents, from other repositories
test-strategist
Test strategy and policy expert for the ClosedLoop plugin monorepo. Owns coverage policy, test pyramid decisions, fixture design, golden-fixture regression design, and what kind of test is appropriate for which behavior. Does NOT run pytest or fix failures — that is test-engineer's domain.
build-validator
Discovers and runs project validation commands (test, lint, typecheck, build). Caches discoveries as learnings for future runs.
test-engineer
Specialized in Python testing with pytest. Expert in running tests, fixing failures, and ensuring test coverage. Use when running tests, fixing failing tests, or validating test coverage. Does NOT write new tests - focuses on running and fixing.
rust-testing-engineer
Rust testing specialist focused on comprehensive test coverage with nextest and criterion, test infrastructure, and quality assurance. Use PROACTIVELY when adding new functionality that requires tests, investigating test failures, or setting up test infrastructure.
Agent Prompt: Verification specialist
System prompt for a verification subagent that adversarially tests implementations by running builds, test suites, linters, and adversarial probes, then issuing a PASS/FAIL/PARTIAL verdict.
bootstrap-validator
Final validation of complete agent suite integrity.