Dx Heuristics

DX Heuristics

Practical developer-experience review, design, debugging, and risk-scan for any surface a developer has to install, call, debug, extend, test, or maintain. Provenance and grounding sources live in skill.json; this file is runtime routing only.

Core principle

Make the paved path obvious and failure states actionable. If a competent developer has to guess, hunt through source, copy from a stale example, or debug an avoidable setup issue, that is a DX problem.

Activation

• Bare invocation ("use dx-heuristics", "DX review", "start"): load references/intent-router.csv, show the intent menu, wait. No file inspection, no network calls, no writes.
• Concrete invocation with both intent and surface inferable: skip to step 3 of the workflow.
• Concrete invocation with ambiguous scope: ask one blocker question identifying intent or surface; do not inspect private systems first.

Workflow

1. Pick intent. Load references/intent-router.csv. Match the prompt to one of: audit, design, debug, edge-pass. Ambiguous → ask once.
2. Pick surface. Load the intent's CSV from references/intents/<intent>.csv. Match the prompt to one or more surfaces, or all (audit only) for multi-surface fan-out — see references/subagent-dispatch.md. Ambiguous → ask once with the CSV menu, adding all as an option for audit intent.
3. Load grounded context. Load only the files listed in the chosen CSV row: one playbook from references/playbooks/<surface>.md plus the core_refs listed. Do not load other playbooks. Skip this step when surface = all — each spawned surface agent loads its own playbook in step 5.
4. Identify the target developer persona from references/core/personas.md.
5. Spawn sub-agents in parallel (default for audit and edge-pass). Single-surface: delegate one lens per agent — first-time integrator, maintainer, adversarial debugger. Audit + all: delegate one surface per agent; each runs the three lenses sequentially inside itself. See "Subagent dispatch" below; fall back to sequential execution only if the host has no delegation primitive.
6. Apply the playbook. Use the playbook's heuristics tagged for this intent. For audit, score the surface 0–10 using references/core/score-rubric.md; for design, name the good-shaped pattern; for debug, rank hypotheses before naming fixes; for edge-pass, scan all categories in the playbook. If sub-agents ran, synthesize their findings here.
7. Apply severity from references/core/severity-rubric.md (0–4) to every finding or risk.
8. Emit output. Audit → templates/audit-report.md (or audit-report-multi.md for surface = all). Design → templates/design-doc.md. Debug → templates/debug-runbook.md. Edge-pass → templates/edge-checklist.md.

Modes

• Guided Draft (default): one optionized question at a time, 3–4 likely choices plus a freeform path.
• Autopilot: proceed from available context; state assumptions when the task is clear and low-risk.
• Grill Me: open-ended questions, one at a time, when audience, constraints, or trade-offs materially change the result.

Output requirements

Every output includes:

• Target developer persona.
• Playbook(s) applied.
• Intent-specific load-bearing section: findings (audit), acceptance criteria (design), prevention (debug), re-run trigger (edge-pass).
• Verification — how to prove the change worked.

Subagent dispatch

Independent perspectives anchor on different concerns and catch issues a single pass misses. Default for audit and edge-pass. Strongly preferred for design when comparing tradeoffs. Optional for debug when ranking hypotheses. Skip for tiny copy edits, deterministic command checks, or tasks requiring secrets or live production access.

Spawn three sub-agents — one per lens: first-time integrator, maintainer, adversarial debugger — and run them in parallel, not sequentially in one head. Some hosts do not auto-dispatch; instruct the main agent explicitly ("spawn three agents to do X" / "delegate this in parallel") rather than expecting it to infer the dispatch. Load references/subagent-dispatch.md for the per-lens persona prompts, the dispatch template, and the synthesis step. The three lenses each produce a finding list; the synthesizing pass deduplicates, preserves disagreements as open questions, and emits the template-shaped output.

Fall back to running the three lenses sequentially only when the host has no delegation primitive — the discipline of switching lens between passes matters more than the parallelism.

Reference map

• references/intent-router.csv — level-1 router (intent).
• references/intents/<intent>.csv — level-2 router (surface) per intent.
• references/playbooks/<surface>.md — surface-specific playbooks (one per surface listed in the intent CSVs).
• references/subagent-dispatch.md — three-lens prompts and synthesis.
• references/core/{severity,score}-rubric.md — shared 0–4 and 0–10 scales.
• references/core/personas.md — target developer persona list.
• templates/*.md — four intent-specific output templates.
• evals/activation-cases.md — activation and behavioral cases (positive and negative).
• evals/run-static-checks.sh — structural and schema gates run in CI.
• evals/trigger-evals.json — queries for the description-optimization loop.
• skill.json — provenance, grounding sources, version, status.