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 skills/bitcraft-apps/spec-first/documentnpx skills add bitcraft-apps/spec-first --skill documentgit clone --depth 1 https://github.com/bitcraft-apps/spec-firstWhat 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 | $0.00008 | $0.00644 |
| Opus 5 | $0.00004 | $0.00322 |
| Sonnet 5 | $0.00002 | $0.00129 |
| Haiku 4.5 | $0.00001 | $0.00064 |
Grade A, and why
document 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 — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Document Command
Creates minimal, proportional documentation. Match doc weight to change weight. One agent does the steps in order.
Usage
/sf:document [SPECIFICATION_AND_IMPLEMENTATION_PATHS]
Context
Read .sf/implementation-summary.md if it exists.
Input Resolution
- If
$ARGUMENTSis non-empty: use it as the artifact and implementation paths - Else if
.sf/spec.mdand.sf/implementation-summary.mdexist: use them - Else: ask the user where the artifacts are. Wait for the answer.
Execution
Write each file in turn. All paths are under $SF_DIR/research/.
artifacts-summary.md— the requirements and outcomes in the artifacts. Read only what exists.implementation-summary.md— the real code structure: the main files, the changed interfaces.docs-inventory.md— every existing doc asfilepath | primary topic, from the headings only.
Gate 1 — Post-Analysis: run scripts/doc-gates.sh analysis, in this skill directory.
If doc-gates.sh fails (non-zero exit), halt immediately — do not write the documents.
technical-docs.md— what a developer needs to use or extend the change. Skip any section that does not apply.user-docs.md— only what the user must do differently. Skip it if the change is invisible.
An empty file is a valid result. It means the change does not need that document.
Gate 2 — Post-Generation: run scripts/doc-gates.sh generation.
If doc-gates.sh fails (non-zero exit), halt immediately — do not integrate.
- Integrate: for each topic in
docs-inventory.mdthat matches, edit that file. Create a file only when none fits. Cut duplicate content. Then delete the research files.
Output: Documentation updates (if any) + terminal summary
Gate Failure Behavior
On a non-zero gate exit: halt, report which gate failed and the reason it printed, and keep the output for inspection.
On Claude Code
Claude Code runs the steps as the sf-document workflow — schema-checked, and with no shared
file between the analysis agents. The workflow applies both gates itself.
What ships with it
1 file 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.
- yesterday First seen · 66 lines · 8 tokens per session scan A e4adb11630ef
document is a skill published in the GitHub repository bitcraft-apps/spec-first (4 stars, last pushed 4d ago), licensed MIT. It adds 8 tokens to every session and 644 once invoked, about $0.0000 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 skills, from other repositories
anti-slop-design
Category-aware design skill that builds distinctive, production-grade UIs. Brand-vs-product register, color strategy commitment, scene-based theme choice, palettes, font pairings, UX patterns, shadcn/token integration, empty-error-loading copy, secondary slop signals, multimodal design parity from mocks, and a…
3d-web-experiences
Build distinctive, performant, production-grade 3D on the web with Three.js, React Three Fiber, and WebGL. Use when the user asks to "build a 3D scene", "add a 3D hero/landing", "make a product viewer/configurator", mentions three.js / react-three-fiber / r3f / drei / webgl / shaders, or asks to make an existing 3D…
deep-research
Conducts multi-step deep research on any topic using iterative search, reflection, and synthesis. Use when the user asks to research, investigate, survey, compare, analyze, deep-dive, or explore a topic in depth. Covers web research, codebase analysis, documentation review, mixed-source investigation, and M3…
minimax-m3-long-context
How to use MiniMax M3's 1M-token MSA context productively: what to load vs. compress, when to retrieve vs. ingest, how to keep skills shallow in the always-on prompt and deep in skills, and how to plan retention across iterations. Load when the task might exceed 200K tokens, when the user asks to "keep all of this in…
minimax-m3-multimodal-input
How to use MiniMax M3's native multimodal input (image, video) for grounded decisions in coding work. Covers reading attached images/frames, treating them as ground truth for visual claims, screenshot diffing, design parity from mockups, and routing visual evidence through reports and PRs. Load when the user attaches…
minimax-multimodal-toolkit
MiniMax-native multimodal workflow for image, video, voice, music, and media-processing tasks. Use when the user asks to generate image/video/audio assets, wants MiniMax-specific media APIs, needs TTS or voice workflows, wants reproducible local media outputs, or needs FFmpeg-style processing around generated media.…