AGENTS.md git:20260723.d48117cA

Immutable. This exact content is served forever at /api/v1/blob/2a4caea1ca8f7bc3.

# AGENTS.md - Hypertaks plugin repo

This repository packages the **Hypertaks Founder** skill as a cross-agent plugin.

- The skill itself lives in [`skills/hypertaks/SKILL.md`](skills/hypertaks/SKILL.md).
  Read it first - it defines the mandatory six-phase loop, Phase 0-5 (sized intake gate →
  frame → pick the tier's roles → equip → produce the agents → integrate &
  deliver) and the Dynamic Agent Allocation tiers (Lite 1 / Standard 3 /
  Prime 5 / Hyper 6–10+).
- Reference material is under `skills/hypertaks/references/` and templates under
  `skills/hypertaks/assets/`.

## Working on this repo

- Keep the skill self-contained; do not add hard dependencies on any single
  agent's tooling.
- All prose stays in **English**.
- When editing the skill, re-validate it (see `.github/workflows/validate.yml`)
  before committing. Retrieval or runtime changes also require the Python
  retrieval tests and `npm test`.
- Manifest files in the dot-directories (`.Codex-plugin`, `.codex-plugin`, etc.)
  are per-agent adapters that all point back to the single `skills/hypertaks`
  skill - keep versions in sync when bumping.
- Keep the Retrieval Intelligence Router in
  `skills/hypertaks/references/02-retrieval-and-evidence.md`, professional
  execution profiles in `03-professional-execution.md`, and visual routing in
  `04-visual-delivery.md`. Do not duplicate their canonical rules across files.
- **One work item, one commit.** W3 and W4 were once squashed into a single
  commit, which made the budget change and the runtime change impossible to
  revert independently. They have since been split. From W5 onward every phase
  lands as its own commit, even when two phases are finished in the same sitting.