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 byheaven/byheaven-skills --skill code-productiongit clone --depth 1 https://github.com/byheaven/byheaven-skillsWrote 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/byheaven/byheaven-skills/code-production)<a href="https://agentmods.dev/skills/byheaven/byheaven-skills/code-production"><img src="https://agentmods.dev/badge/skills/byheaven/byheaven-skills/code-production/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/byheaven/byheaven-skills/code-production"><img src="https://agentmods.dev/badge/skills/byheaven/byheaven-skills/code-production.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 | $0.00038 | $0.01138 |
| Opus 5.5 | $0.00015 | $0.00455 |
| Sonnet 5.5 | $0.00008 | $0.00228 |
| Haiku 4.5 | $0.00004 | $0.00114 |
Grade A, and why
code-production 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 today.
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 — 21 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Code production
Before the first edit, decide whether the change adds or reshapes a type, a module, stored data, or a command or job that writes files or records. If it does, read references/structure.md now; it sets how that new shape is built.
- Make the smallest coherent change that satisfies Acceptance. Modify or delete before adding; remove the superseded path in the same candidate. Prefer targeted edits when they keep the change easy to review; a broader rewrite is appropriate when it produces a simpler reviewable result without absorbing unrelated or concurrent work. The proportion of lines changed does not decide the method.
- In the code you change, a comment carries only what the code cannot say: a license header, behavior forced by a dependency, platform, or protocol we cannot change, a public API contract, or a link to the issue behind a constraint. When our own code needs a comment to explain surprising behavior, reshape the code by renaming, extracting, or typing until it explains itself. A comment that guards a constraint ("do not remove", "keep this wording") stays until a test or check enforces that constraint; when that enforcement is outside owned scope, report it as a follow-up. A lint suppression stays only for a style-only rule; one that silences a correctness or safety rule is resolved in the code.
- Before handing over, read your own diff once as its reviewer and remove what it added without need: comments that narrate the code, defensive checks or try/catch on trusted internal paths, casts that bypass the type checker, and nesting an early return would flatten.
- In a sweep, migration, or batch of similar edits, check each unit before starting the next, so a break shows at the unit that caused it. For such a batch, or a check you will repeat across many units, reuse an existing rerunnable command or script when one covers the work, otherwise write the smallest script that does or proves it, and keep the invocation as a receipt a reviewer can rerun; a couple of edits visible at a glance need none.
- Stay inside owned scope. Surface a required scope expansion instead of silently crossing it. A pre-existing bug, performance concern, or behavior the directive does not mention stays untouched unless Acceptance cannot be met without it; report it as a follow-up. A step, action, entry point, setting, or sentence absent from the approved review object stays out and the gap is reported; within the object's direction, presentation, motion, feedback, and conventional states reach the highest craft the work allows.
- A new test needs an expected outcome from outside the code it checks; a test written from the same reading as the implementation repeats that reading. Three sources supply one. A bug: the fix starts with a case that fails on the unfixed code for the reported reason. A decided requirement that predates the change: the cases are written from its statement by an agent other than the producer, before it sees the candidate change, and arrive in the directive or from a verifier; without such cases the producer adds no requirement test. This independent-author part is on trial. A trust boundary (authentication, money, deletion, privacy, or a guard on production or secrets): independently derived adversarial cases stay required. Every other change adds no test and runs the existing checks that cover it.
- Test behavior through the public interface, at the seams the directive or Intent names. A test that restates the implementation is a defect, not coverage: one that copies a value the code already holds (a count or roster from a registry, copy text, prompt wording, a constant, source or script text), asserts only that a mock was called, or recomputes the expectation the way the implementation does. A check over every entry of a registry enumerates the registry. When an intended change breaks a restating test, delete or rewrite it rather than update its copy. When a module deepens behind a new interface, delete the superseded shallow tests in the same change. Size committed tests like the neighboring test files, roughly one focused test per stated behavior; scratch checks used during verification need not be kept. A committed test runs production code or checks a formal structure such as a schema, path, heading, link, or size; a one-time observation, such as a deleted file being absent, belongs in the result's evidence.
- Run the smallest checks that directly judge the changed predicates. Broaden to regression or build coverage when the actual impact surface or dependency uncertainty requires it, and repeat or broaden passing checks only when new changes, failures, or unresolved concerns justify it. Record exact command, exit or result, and raw output location when available. Silence, timeout, malformed output, or a skipped check is not a pass.
- For a performance change, where speed, latency, memory, size, frame rate, or cost is the target, read references/performance.md before the first edit and follow it through the report.
- Load a domain skill only when the artifact needs that specialist procedure. The skill does not widen scope or change Acceptance.
- A wait you own — a probe still scanning, a check or deploy still running, a process you started — belongs to this change: stay with it until it reaches terminal state or its stated bound, then report.
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- today Changed 48f7ac6153ed
- yesterday Changed · +5 lines eda7f95c3e84
- 9d ago Changed 397eb611c197
- 10d ago Changed b990faba77dc
- 11d ago Changed · +1 lines 9b0429377bc4
- 16d ago First seen · 15 lines · 38 tokens per session scan A f629c9c2f1fe
code-production is a skill published in the GitHub repository byheaven/byheaven-skills (2 stars, last pushed today), licensed MIT. It adds 38 tokens to every session and 1,138 once invoked, about $0.0002 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 2026-10-02.
Other skills, from other repositories
engram-testing-coverage
TDD and coverage standards for Engram. Trigger: When implementing behavior changes in any package.
conductor-implement
Execute tasks from a track's implementation plan following TDD workflow.
principle-test-behavior-not-implementation
Apply when you write, change, or keep a test. Assert what the code's users can observe, so that the test fails when that behavior breaks and keeps passing when only the internals change.
iterative-development
TDD iteration loops using Claude Code Stop hooks - runs tests after each response, feeds failures back automatically.
python
Python development with ruff, mypy, pytest - TDD and type safety.
tdd
This skill should be used when the user wants to implement features or fix bugs using test-driven development. Enforces the RED-GREEN-REFACTOR cycle with vertical slicing, context isolation between test writing and implementation, human checkpoints, and auto-test feedback loops. Uses multi-agent orchestration with the…