Example from HELIX’s own docs. This generated page comes from docs/helix/. Use it to see the method in practice; start with the artifact-type catalog for reusable templates. Historical plans and reports may describe retired architecture.

Source identity (from 02-design/technical-designs/TD-011-slider-autonomy-implementation.md):

ddx:
  id: TD-011
  depends_on:
    - FEAT-011
    - ADR-003
    - helix.prd

TD-011: Slider Autonomy Implementation

Context

[[FEAT-011]] governs the low/medium/high autonomy spectrum. This design specifies how the content-only HELIX realizes that policy: through action prompts, the routing skill, and runtime-neutral artifacts. There is no HELIX runtime to implement — the runtime supplies agency; HELIX supplies the words that tell it how often to pause and what it may never skip.

This replaces the pre-collapse TD-011, which designed a helix input CLI, an execute-loop queue drainer, and a .helix/slider-config.yaml schema. Those surfaces were removed in 823aa1ac and are out of scope. What survives is the behavioral contract.

Requirements summary (from FEAT-011)

1. Three-position spectrum (low/medium/high), default medium (FR-1).
2. Resolution precedence: per-invocation → artifact frontmatter / project policy → runtime default; no CLAUDE.md (FR-2).
3. Concern inference at high autonomy when no concerns.md exists (FR-3).
4. Hard-stop invariant at every level (FR-4).
5. Never-collapse-the-loop invariant at every level (FR-5).

Design decisions

Decision 1 — Autonomy is resolved per action, not stored in a config file

Each action that can pause (input, frame, evolve, design, build) resolves the active level at its bootstrap step using the FR-2 precedence chain:

level = invocation_override            # e.g. "--autonomy high" or explicit prompt instruction
      ?? governing_artifact.frontmatter.autonomy
      ?? project_policy.autonomy        # runtime-neutral project policy artifact
      ?? "medium"                       # runtime default

No file format is mandated beyond a runtime-neutral autonomy: key in frontmatter or a project policy artifact. The level is data the action reads, not a feature it owns. This keeps autonomy expressible identically across DDx, Claude Code, Codex, Copilot, and Genie.

Boundary note: the runtime may surface the per-invocation override however it likes (argument, env var, chat instruction). HELIX only fixes the precedence order and the default, never the runtime mechanism.

Decision 2 — Checkpoint density is the only thing the level changes

An action’s steps are fixed by the action; autonomy parameterizes only the pause points between and within them:

The set of steps the action runs is identical across columns. Only the “ask” cells differ. This is the mechanical statement of FEAT-011 FR-5.

Decision 3 — Concern inference at high autonomy

When an action that touches concerns (frame, input, evolve) runs at high and docs/helix/01-frame/concerns.md is absent:

1. Classify the product nature from the vision/PRD/intent (web app, API service, CLI, data pipeline, library).
2. Map the nature to candidate library concerns under workflows/concerns/ (e.g. web app → a tech-stack concern + a frontend concern + a11y-wcag-aa; API service → tech-stack + o11y-otel + security-owasp).
3. Write concerns.md with the inferred selection, marking each as an assumption (a recorded inference, not a confirmed choice).
4. Detect conflicts among inferred concerns per concern-resolution.md; record unresolved ones as assumptions to revisit, never as a silent pick.

At low/medium, concern selection follows the interactive path in concern-resolution.md §“Concern Selection in /helix frame”. Inference never overwrites an existing concerns.md.

Decision 4 — Hard-stop classification

A hard stop is a condition no autonomy level may pass:

Everything else is a resolvable conflict: high autonomy records an assumption and continues; low/medium ask. A runtime preserve/abort outcome ends the current bounded attempt and returns control to HELIX — it is an execution result, not a hard stop, and must not be re-classified as one.

Decision 5 — Where the behavior is authored

Traceability

Testing strategy

Behavioral, not unit — there is no HELIX code. Verification rides the existing workflow-coverage harness ([[FEAT-014]]):

• A high-autonomy fixture run on a no-concerns.md project produces a concerns.md with inferred concerns (FR-3 / FEAT-011-AC3).
• A high-autonomy fixture with a planted equal-authority contradiction stops rather than picking a side (FR-4 / FEAT-011-AC4).
• An autonomy precedence fixture (invocation override vs frontmatter) resolves to the override (FEAT-011-AC2).
• A loop-collapse check confirms every required activity still runs at high autonomy (FR-5 / FEAT-011-AC5).

Risks and mitigations

References

• [[FEAT-011]] — slider autonomy feature
• [[ADR-003]] — autonomy-spectrum decision
• Input action
• Concern resolution reference