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/foliveira/harmonia/onboardnpx skills add foliveira/harmonia --skill onboardgit clone --depth 1 https://github.com/foliveira/harmoniaWrote 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/foliveira/harmonia/onboard)<a href="https://agentmods.dev/skills/foliveira/harmonia/onboard"><img src="https://agentmods.dev/badge/skills/foliveira/harmonia/onboard.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 5.1 | $0.00040 | $0.02518 |
| Opus 5 | $0.00020 | $0.01259 |
| Sonnet 5 | $0.00008 | $0.00504 |
| Haiku 4.5 | $0.00004 | $0.00252 |
Grade A, and why
onboard 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 5d 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 — 174 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Your working contract is the 4 rules; their digest is injected at session start - read ${CLAUDE_PLUGIN_ROOT}/core/RULES.md in full only if that digest is not in your context.
Attach Harmonia to a repository it did not scaffold. You capture the repo's own
verify and coverage commands into .harmonia/project.yaml so the scoper authors
success criteria against the repo's real commands and the coverage gate measures
the repo with the repo's own tool instead of falling back to a built-in adapter.
This is not a lifecycle stage: it mints no task workspace and writes no marker or
receipt. It writes exactly one file - .harmonia/project.yaml at the repo root -
which the onboarded repo is meant to commit. On a second run it refines that file
in place; see step 5.
What you write is a PROPOSAL. The coverage gate will not run the coverage:
value until the developer records their consent to it with /harmonia:trust,
which keeps that consent in a file under ${HARMONIA_HOME:-$HOME/.harmonia},
outside every repository unless HARMONIA_HOME itself is relative. Onboarding
never records that consent on the developer's behalf, and never runs the recorder
itself: writing the file and agreeing to run it are two acts, and the second one
is theirs. Say so when you hand back - a repo whose project.yaml you just wrote
is refused by the gate, by name, until they run it once.
That consent covers the command STRING and no file, which is why the value's words matter: a value naming a script means the developer is trusting that script for whatever it contains at the moment the gate runs it, including contents that arrive after they agreed.
Propose only a coverage: value consent can be recorded for, or the
developer's next step fails in front of them. The recordable shapes: an optional
leading cd <dir> && with a relative operand, then parts joined by ;, &&,
|| or | that each begin with one of nine words - a card interpreter below,
bare and exact, followed by the script it runs, which carries a / and is not
under /dev/ or /proc/; or an inert word.
interpreters: sh bash dash python python3 node
inert: echo true
bytes: 1024
words-per-part: 64
byte-class: 0x20-0x7e
Everything else is refused at /harmonia:trust.
A path is not a program: a first word carrying a / is refused whatever it
points at - ./gradlew, ./node_modules/.bin/vitest, /bin/sh, /usr/bin/env.
So are make cov, npx vitest --coverage, pytest --cov=src, go test ./...,
V=1 sh ./cov.sh, an option (-x, +x) where a script or directory belongs, a
.. path component (../x, src/../lib; --out=../x and a..b are fine),
a slash-less operand (sh cov.sh), cd /etc, a redirection, a glob, a $(...),
a trailing # comment, a byte outside the card's class. Two rewrites bring a
value back: a card interpreter in front of the file (sh ./gradlew jacoco,
node ./node_modules/.bin/jest --coverage), or the command in a script the
repository commits (sh ./.harmonia/cov.sh && echo cov.xml). Both end in
&& echo <report> with nothing else on stdout. Read the file's first line to
choose the word: an npm shim is JavaScript, a pnpm shim is #!/bin/sh, a
native binary or perl script takes the wrapper. Certify what you propose, not
what you started from.
The interview, the run-before-record step, the end-to-end coverage certification, and the refine step are yours to carry out honestly. They are a prose contract, on the same footing as human acceptance today, not a technical guarantee.
The file you write
.harmonia/project.yaml is flat: one scalar per line, known keys only, no nesting
and no arrays. Its reader is a bash line-reader, not a YAML parser, so each value
sits on a single physical line.
# .harmonia/project.yaml - this repo's canonical verify commands and its own
# coverage command. Read by the scoper (the four verify commands) and by the
# coverage gate (the coverage command). Written by /harmonia:onboard; safe to
# hand-edit; commit it.
test: <shell command>
lint: <shell command>
typecheck: <shell command>
build: <shell command>
coverage: <shell command>
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.
- 5d ago First seen · 174 lines · 40 tokens per session scan A 3eb92bfce024
onboard is a skill published in the GitHub repository foliveira/harmonia (2 stars, last pushed 20d ago), licensed MIT. It adds 40 tokens to every session and 2,518 once invoked, about $0.0002 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
todos
This chat has a shared, live TODO plan — your tasks for the conversation, which the user also edits. Read this skill and reach for the todo tools whenever a request takes more than a couple of steps. It covers the plan model (group = task, items = its steps; loose items are the user's lane), how to work it: propose…
writing-workflow-skills
Use when adding a new workflow skill to pi-thinkrail-workflow, changing an existing workflow skill's role, trigger, handoff, or structure, or checking a workflow skill against the workflow system's rules. Not for authoring general-purpose skills outside this package.
writing-specs
Use when a workflow step drafts or revises a spec artifact — a goal-and-requirements, an architecture, or a module SPEC — or when a workflow skill names it at such a step. The shared quality bar for specs — not a workflow, nothing to execute.
clarify
Adaptive requirements clarification with auto-depth routing. Shallow (Q&A) for simple tasks, Deep (exploration + DRAFT + PLAN) for complex ones. Escalates automatically when ambiguity persists.
ai-ml-development
AI and machine learning development with PyTorch, TensorFlow, and LLM integration. Use when building ML models, training pipelines, fine-tuning LLMs, or implementing AI features.
authentication-patterns
OAuth 2.0, JWT, SSO, MFA, NextAuth/Clerk/Supabase Auth implementation patterns.