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 shennawardana23/skillme --skill agent-harness-constructiongit clone --depth 1 https://github.com/shennawardana23/skillmeWrote 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/shennawardana23/skillme/agent-harness-construction)<a href="https://agentmods.dev/skills/shennawardana23/skillme/agent-harness-construction"><img src="https://agentmods.dev/badge/skills/shennawardana23/skillme/agent-harness-construction/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/shennawardana23/skillme/agent-harness-construction"><img src="https://agentmods.dev/badge/skills/shennawardana23/skillme/agent-harness-construction.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 | $0.00095 | $0.01839 |
| Opus 5 | $0.00048 | $0.00920 |
| Sonnet 5 | $0.00019 | $0.00368 |
| Haiku 4.5 | $0.00010 | $0.00184 |
Grade A, and why
agent-harness-construction 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 11d 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 — 175 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Harness Construction
The "harness" is everything between the model and the real system: tool schemas, the text the model sees back from a tool call, and the rules for when to retry versus stop. Agent quality is bounded by this interface as much as by the underlying model — a capable model given ambiguous tools or opaque errors will still loop, guess, or give up.
The four levers
- Action space — which tools exist, how they're named, how narrow their inputs are.
- Observation shape — what comes back from a tool call.
- Recovery contract — what the model is told to do when a call fails.
- Context budget — how much of the harness's own guidance and history the model has to hold in its context window at once.
Treat these as one design surface, not four independent decisions — a vague action space usually shows up downstream as a recovery problem ("the model called the wrong tool, got a cryptic error, and retried the same wrong call").
Action space design
- Stable, explicit names.
create_reservation, nothandle_request. A name should tell the model when not to use it as much as when to. - Schema-first, narrow inputs. Prefer several small typed parameters over one free-text field the model has to format correctly from memory. Every extra way to phrase the same input is a way for the model to get it wrong.
- Deterministic output shape. The same tool call against the same state should return the same shape every time — varying which fields are present based on a hidden internal state forces the model to guess.
- Avoid catch-all tools (
run_command,execute_sql) unless there is genuinely no way to scope the operation — a catch-all tool pushes all validation and safety judgment onto the model at call time, which is exactly the point where mistakes are hardest to catch.
Granularity
Match tool grain to risk and frequency, not to what's convenient to implement:
| Grain | Use for | Example |
|---|---|---|
| Micro | High-risk, hard-to-undo operations | deploy_to_production, run_migration, grant_permission |
| Medium | The common read/edit/search loop | read_file, search_code, apply_patch |
| Macro | Only when round-trip latency dominates cost | run_full_test_suite |
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.
- 11d ago First seen · 175 lines · 95 tokens per session scan A 59029f18503a
agent-harness-construction is a skill published in the GitHub repository shennawardana23/skillme (2 stars, last pushed 13d ago), licensed Apache-2.0. It adds 95 tokens to every session and 1,839 once invoked, about $0.0005 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
writing-skills
Use when creating new skills, editing existing skills, or verifying skills work before deployment.
receiving-code-review
Use when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation.
writing-plans
Use when you have a spec or requirements for a multi-step task, before touching code.
skill-creator
Create, improve, evaluate, benchmark skills. Use when authoring a new skill, updating an existing one, running evals, or optimizing a skill's description for triggering. Don't use for invoking skills, writing prose, or scaffolding Python projects.
skill-index-updater
Add GitHub skill repos to the ASM index: clone, audit, eval, regenerate index, rebuild catalog, open PR. Use when given GitHub URLs to onboard. Don't use for authoring (skill-creator), improving (skill-auto-improver), or install (asm install).
hello-world
A minimal test skill that greets the user and demonstrates the ASM publish workflow.