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/harnessprotocol/harness-kit/harness-compilenpx skills add harnessprotocol/harness-kit --skill harness-compilegit clone --depth 1 https://github.com/harnessprotocol/harness-kitWrote 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/harnessprotocol/harness-kit/harness-compile)<a href="https://agentmods.dev/skills/harnessprotocol/harness-kit/harness-compile"><img src="https://agentmods.dev/badge/skills/harnessprotocol/harness-kit/harness-compile.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 | $0.00108 | $0.04970 |
| Opus 5 | $0.00054 | $0.02485 |
| Sonnet 5 | $0.00022 | $0.00994 |
| Haiku 4.5 | $0.00011 | $0.00497 |
Grade B, and why
harness-compile scanned grade B with 1 finding 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 4d 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
**Codex and Windsurf do not get a project-level MCP file written.** Codex configures MCP servers via its own `~/.codex/config.toml`, a TOML format this compiler doesn't generate; Windsurf's MCP config is global (`~/.code How it starts
The opening of the file, as written. The whole thing — 398 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Compile a Harness Configuration
You are compiling a harness.yaml file into native configuration files for one or more AI coding tools. Each target tool gets its own idiomatic output: instruction files, MCP server configs, skill files, and permission settings.
This skill implements the Harness Protocol compiler mapping defined at harnessprotocol.ai. The output is deterministic — the same harness.yaml always produces the same files.
The 8 supported targets: claude-code, cursor, copilot, codex, opencode, windsurf, gemini, junie. The last five (Codex, OpenCode, Windsurf, Gemini CLI, Junie) all read operational instructions from a single shared AGENTS.md file at the project root — see Step 3 for how that sharing works.
Workflow Order (MANDATORY)
Follow these steps in order. Do not skip any step.
Step 1: Find harness.yaml
Check for the harness file in this order:
- A path provided by the user after
/harness-compile(e.g.,/harness-compile ~/dotfiles/harness.yaml) ./harness.yamlin the current directory
If no file is found at either location, tell the user:
"No
harness.yamlfound. Specify a path:/harness-compile path/to/harness.yaml"
Read and parse the file. Extract metadata.name (use default if absent). Extract version to confirm v1 format. Note the instructions.import-mode value (default: merge if not specified).
If version is absent, is the integer 1 (legacy format), or is any value other than the string "1", stop and tell the user: "This harness.yaml uses an unsupported format. Run /harness-validate to check the file and see upgrade instructions."
Also check for any flags the user passed:
--target <targets>— comma-separated subset ofclaude-code,cursor,copilot,codex,opencode,windsurf,gemini,junie, orallfor every target--dry-run— preview output without writing any files--clean— remove orphaned marker blocks in addition to current update--verbose— show skipped slots and extra detail in the compilation report
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.
- 4d ago First seen · 398 lines · 108 tokens per session scan B 62017ed1d1b5
harness-compile is a skill published in the GitHub repository harnessprotocol/harness-kit (10 stars, last pushed today), licensed Apache-2.0. It adds 108 tokens to every session and 4,970 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
llm-wiki
Build and maintain an LLM-curated personal knowledge base — the "LLM Wiki" pattern from Andrej Karpathy's April 2026 gist. Use this skill whenever the user wants to ingest a source (paper, article, transcript, PDF, notes) into a persistent compounding knowledge base, ask a question against accumulated notes, lint or…
studio
Architecture Studio control plane — initialize or inspect a studio workspace, create and register projects, or route an architecture/AEC task to the right agent or skill. Use when the user runs /as:studio, asks to set up or open their studio, manage its projects, or describes a task without naming a skill.
product-image-processor
Download, resize, and remove backgrounds from product images at scale. Use when the user asks to "process product images", batch-download images from the schedule, strip backgrounds, or standardize product photos.
tool-catalog
Show the available Architecture Studio skills and their host-appropriate commands, plus Claude Code's native Architecture Studio agents. Use when the user invokes tool-catalog, asks "what can you do" or "what skills are available", or wants a directory of plugin commands. Do not use for a host's native skill-list…
cesiumjs-imagery
CesiumJS imagery layers - ImageryProvider, ImageryLayer, ImageryLayerCollection, WMS, WMTS, Bing, OpenStreetMap, ArcGIS, Mapbox, tile discard policies. Use when adding or swapping base map layers, configuring imagery providers, layering multiple map sources, or creating split-screen imagery comparisons.
studio-feedback
Prepare a user-reviewed Architecture Studio bug report or feature request for GitHub. Use when the user runs /as:studio-feedback, says "report a bug", "this skill broke", "send feedback", or requests a feature. Do not use for Claude Code's native /feedback command.