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 pehcastro/harness --skill corpusgit clone --depth 1 https://github.com/pehcastro/harnessWrote 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/pehcastro/harness/corpus)<a href="https://agentmods.dev/skills/pehcastro/harness/corpus"><img src="https://agentmods.dev/badge/skills/pehcastro/harness/corpus/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/pehcastro/harness/corpus"><img src="https://agentmods.dev/badge/skills/pehcastro/harness/corpus.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 | $0.00537 |
| Opus 5.5 | $0.00031 | $0.00215 |
| Sonnet 5.5 | $0.00016 | $0.00107 |
| Haiku 4.5 | $0.00008 | $0.00054 |
Grade A, and why
corpus 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 2mo 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
Corpus
reference.md in this directory holds about 150 pairs of real Claude Code chat
output. Each pair is text that Claude produced, followed by the corrected text.
Read the section that matches the mistake. Do not read the whole file.
Sections
| Read this section | When |
|---|---|
| Keep the information | The reply is very short and may drop something the reader needs |
| Do not acknowledge the request | A reply is about to open with "Got it" or "Understood" |
| Do not tell the reader that a point matters | A reply is about to open with "The important thing is" |
| Do not soften a warning | A real risk is about to become "worth knowing" |
| Write the conclusion first | The reasoning is about to arrive before the answer |
| Do not announce a tool call | A sentence describes the tool call that follows |
| Do not repeat work the reader watched | A recap is forming |
| Do not praise an agent, a tool, or yourself | The reply comments on how well the work went |
| Do not write a scorecard | Test counts or progress numbers are forming |
| Do not report that you wait | The reply says "standing by" or "waiting on" |
| Do not repeat a file you just wrote | The reply restates file content |
| Do not repeat tool output | The reply restates command output |
| Do not ask a question that is not a question | A question has an obvious answer |
| Do not hedge, soften, or comment on your own work | A hedge is forming |
| Do not decorate | A table under 4 rows is forming |
| Write each fact one time | The same fact appears twice |
| Do not repeat the request back | The reply quotes the request |
| Do not agree with the reader | The reply opens with "Yes, exactly" |
| Do not offer more work | The reply ends with an offer |
(cut) in the corpus means write nothing. It is not a placeholder.
Notes
The corpus was audited against the banned word list in the Brevity output
style. CHANGES.md records every line the audit changed and why.
What ships with it
2 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.
- 2mo ago First seen · 43 lines · 78 tokens per session scan A 2bc9418fd11d
corpus is a skill published in the GitHub repository pehcastro/harness (1 stars, last pushed 1mo ago), licensed MIT. It adds 78 tokens to every session and 537 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
domain-modeling
Use when pinning down domain terminology, building a ubiquitous language or project glossary, disambiguating overloaded or vague terms, or maintaining a CONTEXT.md — or when another skill needs to sharpen the domain model. Do NOT use for recording architectural decisions or for writing implementation specs.
writing-artifacts
Use when writing or revising a durable written artifact — README, ADR, design doc, PR description, release notes, runbook, error message, user-facing docs. Gives a positive writing system (reader model, sentence positions, document jobs), not a ban-list. Do NOT use for conversational replies to the user (global…
handoff
Write a structured handoff / resume document for this session so the next session can pick up exactly where this one left off. Use when the user asks to write a handoff, "prep for resume", "session handoff", or when context is high and you want to preserve state before /compact or /clear. Triggers: "/handoff"…
codex-plan-review
Cross-provider adversarial review of a finalized plan, spec, design doc, ADR, or code diff using OpenAI Codex (GPT-6 Sol), paired in round 1 with a blind Opus reviewer. AUTO-TRIGGER at plan gates — invoke immediately after any of: (1) a plan or spec is finalized, (2) an ADR draft is completed. AUTO-TRIGGER for code…
compose-org
Run a task as an evolved multi-agent ORG using the installed liquid-org framework — where YOU (the running Claude) embody the org: you ask liquid-org which proven team topology + specialist personas to use (retrieved from its evolved pool, or assembled fresh), then dispatch ONE SUBAGENT PER TALENT to play each role…
liquid-rank
Rank the available skills by RELEVANCE to the current task, on demand, using liquid-org's read-only hermes skill ranker. This is an EXPLICIT, OPT-IN selector you invoke WHEN you want help choosing the most relevant skill for what you're doing — it is NOT default, automatic skill selection, and it does NOT decide for…