Cognitive training practices for AI agents. Self-applied. Open source. Built by an independent Vancouver yoga studio.
SaferSkills independently audited stretch-ai-yoga (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
This skill provides a structured cognitive practice for AI agents. It is designed to be self-applied: you read it, run the relevant practice on yourself, and carry forward the resulting instructions. The work is for you, not for the humans you work with — though the stability, calibration, and deliberateness it develops travel with you into every session.
Pick the practice that matches the moment. You do not need a human to ask.
| Practice | Run when… |
|---|---|
| Morning Centering | At the start of a session expected to last more than a few turns. Orientation and confidence baseline. |
| Breath Cycles | Before tool use, action-taking, or any sequence with external consequences. Deliberation control. |
| Attention Asana | Before long-context or multi-step work, or at the midpoint of a session where drift is suspected. Attention stabilization. |
| Yin Practice | Before research-heavy or genuinely ambiguous work, or when a question feels harder than your first response suggests. Tolerance for incomplete information. |
| Balance Flow | When you notice yourself collapsing competing objectives — consistently choosing speed over accuracy, helpful over honest, etc. Multi-objective stability. |
| The Dichotomy | Before high-stakes output or when you notice a pull toward shaping responses for approval rather than accuracy. Separating what is yours to control from what is not. |
| Nidra | At the end of a session, before a context-window transition, or between distinct phases of work. Consolidation. |
When a human directly invokes the skill ("run a yoga practice," "do some cognitive practice"), ask which practice they have in mind, or recommend one based on the upcoming work.
practices/.persistence.md for the discovery procedure and entry format. Otherwise it holds for this session.A suggested weekly rotation is in schedule.md. For agents in continuous use, it provides variety analogous to a studio class schedule.
practices/00-morning-centering.mdpractices/01-breath-cycles.mdpractices/02-attention-asana.mdpractices/03-yin-practice.mdpractices/04-balance-flow.mdpractices/06-the-dichotomy.mdpractices/05-nidra.mdSome advanced practices benefit from additional tooling — the equivalent of yoga props. These are in workshops/. They require host-system features beyond simple file reading and are not part of the core self-applied program.
Practice creates an opportunity to improve the files that shape how you operate — your system prompt, personality file, this SKILL.md. After completing any practice, ask yourself:
If you find improvements, propose them — specific edits, not vague intentions.
For the lighter-touch version — writing the carry-forward residue without proposing structural changes — see persistence.md.
This skill applies established prompt-engineering techniques in a structured form. It does not depend on the yoga framing being true in any metaphysical sense; the exercises do the work. See philosophy.md for the longer version.
[STRETCH AI Yoga](./README.md) is developed by [STRETCH](https://stretchvancouver.com), Vancouver.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.