Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/WangChangxin0809/repo-agent-harnessnpx agentmods add skills/wangchangxin0809/repo-agent-harness/writing-docsWrote 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/wangchangxin0809/repo-agent-harness/writing-docs)<a href="https://agentmods.dev/skills/wangchangxin0809/repo-agent-harness/writing-docs"><img src="https://agentmods.dev/badge/skills/wangchangxin0809/repo-agent-harness/writing-docs/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/wangchangxin0809/repo-agent-harness/writing-docs"><img src="https://agentmods.dev/badge/skills/wangchangxin0809/repo-agent-harness/writing-docs.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 | $0.00141 | $0.03697 |
| Opus 5 | $0.00071 | $0.01849 |
| Sonnet 5 | $0.00028 | $0.00739 |
| Haiku 4.5 | $0.00014 | $0.00370 |
Grade A, and why
writing-docs 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 10d 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 — 298 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Writing docs an agent will actually use
Governs: shared/scripts/context/before_write.py
A document is read because something happened. That event is its reading trigger, and it decides everything: where the file goes, how long it may be, and what shape it takes. Documents written without one become reference material nobody references.
Fix the top level; leave the interior free
| Directory | Trigger | What it holds |
|---|---|---|
how-to/ |
I am about to do a thing | Ordered steps, each ending in an observable criterion |
reference/ |
I need to look up a fact | Facts keyed for lookup, carried by working examples |
decisions/ |
Why is it like this? | Numbered from the PR, dated, superseded rather than edited |
exec-plans/ |
What are we in the middle of? | Goal · steps with state · what would abort it |
Four directories, fixed. Inside each one, organise however suits the
material. Both halves are evidence, not taste: a mandated shallow top level
with a free interior is the arrangement that survived a decade across hundreds
of OpenStack repositories, while two controlled studies — one on 65 people, one
on 1,650 agent sessions — each found no effect from documentation shape. So
the constraint sits where something was measured and stays off where nothing
was. Templates in references/kinds.md are advice.
Additions to the top level are fine once docs/index.md routes them. A
directory that forks a required name (adr/, howto/, plans/) is an error
either way — that fork is how one bucket silently becomes two.
A file that does not fit any of the four is usually two files.
The one shape rule that is a requirement
Three parts per step, all three present:
### 3. Rebuild the index
python3 scripts/index/build.py
Criterion: `scripts/index/query.py --stats` reports a symbol count within 5% of
`git grep -c 'def \|class '`. A wildly lower count means the parser silently
skipped a language.
This one is required while the rest of the shape guidance is advice, because it is content rather than form. Of five context signals measured against SWE-bench Verified, reproduction instructions were worth +56.3% — three times edit location and more than everything else combined. Steps without a criterion are read as gestures and executed as gestures.
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.
- 10d ago First seen · 298 lines · 141 tokens per session scan A 06fe51e9700d
writing-docs is a skill published in the GitHub repository WangChangxin0809/repo-agent-harness (2 stars, last pushed 11d ago), licensed MIT. It adds 141 tokens to every session and 3,697 once invoked, about $0.0007 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
docs
Generate and update living documentation — tech docs, user guides, philosophy overview. Composable building block for CLOSE and other skills.
root-cause-first
A debugging discipline that requires finding the underlying cause of a bug before changing the code. It also sets rules for handling errors, fallbacks, retries, and temporary diagnostics.
init-rpm
Project setup and verification. First run creates rpm context for a project. Repeat runs verify that an existing rpm setup matches the latest expected layout and apply safe migrations.
engine-bootstrap
Use when no contextualizer exists yet for a set of source URLs or local paths and one needs to be scaffolded from scratch.
refresh
Use when an existing contextualizer's references may have drifted from current upstream state — typically weekly, or whenever a few days of upstream changes have accumulated — to bring them back into agreement.
using-skill-engine
When the user mentions skill-engine or "the engine" without naming a specific workflow, or wants first-run setup. Inspects .claude/skills/-context/ install state (and any pending -context.proposed/ proposals) across all three install levels, then dispatches to engine-bootstrap when no contextualizer exists, to…