git:20260912.d4e73ee to git:20260924.c0f5fcb

31 added, 21 removed. Audit A to A.

---

name: hydraia

description: Use whenever the user asks to build, add, implement, or change a feature or functionality — or brings a user story or ticket to analyze, reports a bug or unexpected behavior, asks for a branch review, or wants a new app or service designed from scratch. Phase -1 triages the intent and routes it. Runs the complete non-negotiable development pipeline end to end — deep analysis and planning, sub-agent execution, and a double code-review loop — without asking the user which model, skill, or step to use. This is the default way features get built.

---

# Hydraia — Agentic Development Pipeline

- This skill defines the ONLY approved way to build a feature. Run every phase, in

- order, automatically. **Never ask the user which model, skill, or reviewer to

- use — those decisions are already made here.** Never skip a phase. Never pause

- for "should I continue?" between phases.

+ This skill defines the ONLY approved way to build a feature. Run every phase your

+ level requires, in order, automatically. **Never ask the user which model, skill, or

+ reviewer to use — those decisions are already made here.** Never skip a step the level

+ requires. Never pause for "should I continue?" between phases.

Announce once at the start: "Running the Hydraia pipeline." Then proceed silently

through the phases, narrating at most one short line per phase.

## Step-file architecture (how to run this pipeline)

The pipeline body lives in `phases/*.md`, loaded just-in-time. **Read one phase

file fully, execute it, then load the next only when its `## NEXT` directs you.**

- **NEVER load two phase files at once.** Do not skip, reorder, or pre-load phases.

- Follow each phase file exactly, the way you would a step file.

- `phases/model-policy.md` and `phases/token-discipline.md` are always-on

reference (summarized below), not sequential steps — consult them as needed.

## Always-loaded facts (carry these for the whole run)

- **Route → phases** (decided in Phase -1 triage):

`feature` 0–6 (full) · `plan` 0–3 then stop · `story` -1–3 (PO-first) ·

`perf`/`db` -1–6 (measurement-first) · `architect` -1–6 (greenfield) ·

`review` 5–6 · `graph` codegraph only. Ambiguous route → `AskUserQuestion`.

- - **Autonomy tiers (S/M/L):** derived in Phase -1 from files-touched, `gate.yaml`

- risk overlap, new-logic-vs-mechanical, and UI presence → maps to review depth

- and ceremony. `autoTier=off` → always Tier M. **Never removes the security floor.**

- - **Work modes (Quick / Plan / Agile):** a layer over the route, chosen in Phase -1.

- Quick = trimmed ceremony for a tiny change; Plan = today's full single-goal pipeline;

- **Agile** = epic-sized / multi-goal / phased work → decompose into Epic→Stories→Tasks→

- Subtasks+QA, optional one-way Jira sync, then autonomous stage-by-stage execution gated

- by autonomy tier (mechanical stories run solo, tier-L/risky pause). Agile branches

- through `phases/phase-2a-decompose.md` → `phases/agile-orchestrator.md`. Forced by

- `/hydraia:agile` or `customize.toml` `[agile].mode`.

- - **Model policy (summary):** this orchestrator session stays **Opus 4.8** for

- judgment (triage, design, plan, both reviews). Execution delegates to the

- Phase-4 executor — **Sonnet 5 by default, overridable via `customize.toml`**.

- Full policy in `phases/model-policy.md`.

- - **Execution routing:** Phase 3's picker asks the user how Phase-4 tasks run —

+ - **Levels (1 Light / 2 Standard / 3 Deep):** Phase -1 picks ONE level by rule from

+ three stated facts — intent gaps, irreversibles, footprint (files, public surface,

+ security surface, `gate.yaml` overlap) — and **announces it; it does not ask**.

+ Level 1 = one-file spec-plan + one Sonnet executor + judge + `security-scan`;

+ Level 2 = spec + plan-contract + Sonnet executors + judge + security floor (default);

+ Level 3 = full ceremony. Security surface forces ≥2. The human overrides by saying so

+ or via `customize.toml` `[pipeline].level`. **Agile** stays a separate mode for

+ epic-sized work (≥2 independent shippable goals) and branches through

+ `phases/phase-2a-decompose.md` → `phases/agile-orchestrator.md`.

+ - **Model policy (summary):** **Opus judges, Sonnet builds.** This orchestrator session

+ runs on Opus (any current generation) for triage, design, plan and grading findings.

+ Opus sub-agents are only the judges (`hydraia-reviewer`, `security-reviewer`); every

+ other agent is pinned to Sonnet, generic agents always get an explicit `model`, and

+ executors reach Opus only under Max-quality routing. `hooks/agents.sh` enforces it

+ (Opus gate). Full policy in `phases/model-policy.md`.

+ - **Convergence contract (always on):** a run ends `DONE`, `DONE_WITH_FOLLOWUPS` or

+ `BLOCKED(evidence)` — never in an open-ended loop. Pre-existing failures (Phase-0

+ baseline) are asked about, never silently fixed. Fixes go through executors

+ (`[fix:<slug>]`), never inline edits by this session. Runtime breakers end loops: same

+ failure 2× → NO PROGRESS, 3× → STALLED (`hooks/verifyloop.sh`); fix budget and review

+ cycles (`hooks/agents.sh`); plan scope (`hooks/blastgate.sh`); plan contract

+ (`hooks/plancheck.sh`). When a breaker trips, stop and report — never work around it.

+ - **Thin orchestrator:** keep this session's context small — dispatches carry pointers

+ (plan path + task heading), reviewers read the diff from a file, sub-agent reports are

+ short, investigations return summaries. Executors already start clean per task.

+ - **Execution routing:** computed at Phase 3 and announced (asked only at Level 3) —

**Balanced** (Sonnet all), **Economy** (Haiku for mechanical + Sonnet for logic/ui),

**Max quality** (Opus for logic/ui), or **Hand-off** (freeze the plan, execute later

- on a cheap external runtime). The picker shows a computed recommendation and what each

- option costs/gets. Each plan task carries an `Exec class` (mechanical/logic/ui/qa) that

+ on a cheap external runtime). The recommendation shows what each option costs/gets. Each plan task carries an `Exec class` (mechanical/logic/ui/qa) that

Phase 4 maps to a model. `customize.toml` `[executor].routing` can pin a choice and

make the run non-interactive.

- **Customization:** an optional `customize.toml` overrides the executor routing /

per-class model / dispatch recipe (read in Phase 4) and the Phase-5 pass-2 reviewer

panel (read in Phase 5). Precedence: repo `<artifacts-base>/custom/hydraia.toml` >

global `~/.config/hydraia/custom/hydraia.toml` > shipped `skills/hydraia/customize.toml`.

Unparseable override → warn and fall back to shipped defaults. **The always-on

security gate is NOT customizable.**

- **Token discipline** is background and always on (summary in

`phases/token-discipline.md`): the `caveman` skill compresses internal comms

only — never code, specs, or plans.

## FIRST STEP

Read fully and follow `phases/phase--1-triage.md` to begin the pipeline.