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 skills add MCKRUZ/claude-code-sdlc --skill eval-buildergit clone --depth 1 https://github.com/MCKRUZ/claude-code-sdlcWrote 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/skills/mckruz/claude-code-sdlc/eval-builder)<a href="https://agentmods.dev/skills/mckruz/claude-code-sdlc/eval-builder"><img src="https://agentmods.dev/badge/skills/mckruz/claude-code-sdlc/eval-builder/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/mckruz/claude-code-sdlc/eval-builder"><img src="https://agentmods.dev/badge/skills/mckruz/claude-code-sdlc/eval-builder.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00078 | $0.00658 |
| Opus 5.5 | $0.00031 | $0.00263 |
| Sonnet 5.5 | $0.00016 | $0.00132 |
| Haiku 4.5 | $0.00008 | $0.00066 |
Grade A, and why
eval-builder 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 1mo 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.
How it starts
The opening of the file, as written. The whole thing — 44 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Eval builder
For LLM-powered work, "tests pass" is not sufficient verification. The golden set is the spec's acceptance criteria; CI runs it like tests, and a regression in it blocks a merge like a failing test. This skill builds one that actually tracks quality.
Procedure (grounded in Anthropic's agent-eval guidance)
- Start from real failures, small. 20–50 tasks is a strong start — drawn from the manual pre-release checks and the bug/support queue, not invented in the abstract. Each task is one a second person would grade the same way (unambiguous, with a reference answer).
- Pick graders deterministic-first. Prefer a
state_check(did it reach the right end state?) ortranscript_constraint(e.g. finished in ≤ N turns) over an LLM judge. Use anllm_rubriconly for what rules can't capture (tone, explanation quality), and grade the output, not the path — don't assert an exact tool-call sequence; agents find valid routes you didn't predict. - Compose multidimensional success where needed (state + transcript + rubric), with partial credit for multi-part tasks instead of all-or-nothing.
- Write it under
eval-datasets/specs/<feature>/, referencing the spec file, versioned:eval-datasets/specs/<feature>/golden-set.yamlwithspec: specs/NNNN-<feature>.md(usekit/eval-datasets/golden-set.template.yamlas the shape). Set the threshold the spec requires (e.g. ">= 95%"). - Plan for variance. Agents are stochastic — the suite runs multiple trials per task from a clean, isolated environment (so prior-trial state can't leak or be gamed). Note the trial count.
Calibrate before you gate
- Validate the LLM judges against a few human-graded cases before trusting them; the regression trip-wire (~±3% is a practitioner starting point, not a constant) must be calibrated to the suite's measured variance before it becomes a required check.
Done when
- A versioned
golden-set.yamlexists undereval-datasets/specs/<feature>/(referencing the spec file) with 20–50 grounded cases. - Graders are deterministic where possible; LLM rubrics grade output, not trajectory.
- The threshold and trial count are set; the judge has been sanity-checked against human grades.
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.
- 1mo ago First seen · 44 lines · 78 tokens per session scan A 08683794e760
eval-builder is a skill published in the GitHub repository MCKRUZ/claude-code-sdlc (8 stars, last pushed 10d ago), licensed MIT. It adds 78 tokens to every session and 658 once invoked, about $0.0003 per session on Opus 5.5. 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 skills, from other repositories
research-engineer
An uncompromising Academic Research Engineer. Operates with absolute scientific rigor, objective criticism, and zero flair. Focuses on theoretical correctness, formal verification, and optimal implementation across any required technology.
eval-dataset-design
A guide to creating reliable test sets for AI agents. Each evaluation task includes the request, the simulated user's behaviour, a reset starting state, and a success check that can be independently verified.
tika-eval-compare
Compare extracts from two Tika builds over a corpus to detect regressions in content, encoding, exceptions, and embedded-document handling. Use for "compare before/after extracts", "eval this change against the corpus".
cua-driver
Use Cua Driver for desktop or browser tasks that are awkward or unavailable through Bash/APIs, or when the user explicitly wants GUI interaction: app testing, visual bug reproduction, form filling, calendar entry, screenshots, and demo recording. Also covers Cua setup; not OpenAI Codex Computer Use or web research.
jetson-validate-image
Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both. Not for build or flash steps. Triggers: validate bsp, on-target validation.
atmos-validation
Validate Atmos projects, components, arbitrary JSON Schema inputs, EditorConfig, and GitHub Actions; use affected-file selection and native CI annotations.