메뉴
Classifies GitHub issues/PRs into PR-type categories and recommends autor techniques. Used by packages/core/src/decomposer.ts.
Multi-PR triage and parallel work dispatcher. Prevents single-PR tunnel vision by enforcing a survey-before-deep-dive protocol.
Use this skill when working in repositories managed by Agent Orchestrator or when the user asks how to use `ao` properly. Covers the default AO workflow: bootstrap with `ao start`, dispatch work with `ao spawn`, inspect progress with `ao status` or `ao session ls`, steer sessions with `ao send`, and recover or clean up sessions safely.
TDD-driven evidence workflow for generating authoritative failure/fix proof in PRs.
Canonical 7-green PR merge criteria, PR status check pattern, PR freeze discipline, and admin merge protocol
Define and run skeptic exit criteria for non-trivial tasks — independent verification agent with inverted incentive to find gaps
How to record asciinema/tmux evidence videos that prove work was done correctly
| name | technique-router |
| description | Classifies GitHub issues/PRs into PR-type categories and recommends autor techniques. Used by packages/core/src/decomposer.ts. |
Classifies a GitHub issue/PR description into a PR-type category, then recommends the appropriate autor technique. This is classification-only until routing is wired: classifyPrType() returns PR-type metadata, but technique resolution still happens separately through prompt/config defaults. All classification is delegated to the model API — no heuristic scoring, keyword matching, or hardcoded routing tables.
| Type | Description |
|---|---|
state-bool | Widening boolean semantics (int 1/0, string '1'/'0' as valid True/False) |
data-norm | Normalizing key names and numeric formats (xp→xp_gained, NaN handling) |
ci-workflow | GitHub Actions workflow changes (YAML scripting, gate logic) |
typeddict-schema | TypedDict + validation for previously untyped data structures |
large-arch-refactor | Module extraction, moving functions between files |
unknown | Cannot determine from the description |
All 9 autor techniques converge within rubric noise (~80-85 confidence-interval overlap). SR-prtype is the safe default for all PR types. No per-type routing is statistically justified.
| Technique | Description | When to use |
|---|---|---|
SR-prtype | Self-Refine with PR-type classification | Safe default for all types |
SR-fewshot | Self-Refine with single prior PR exemplar | When a highly similar prior PR exists |
SR | Basic 3-round self-refine | Fallback |
ET | Extended chain-of-thought reasoning | Complex architectural decisions |
PRM | Process Reward Model (score each step) | Debugging, multi-step reasoning |
/technique-router <issue-title> <issue-body-or-url>
This skill does NOT:
This skill DOES:
Used by packages/core/src/decomposer.ts's classifyPrType() function in the AO core library. That function classifies PR type only; it does not yet select or apply a technique directly.