skills/deadcode/SKILL.md
Find and delete dead code — zero callers means zero reasons to exist
npx skillsauth add api-haus/my-claude-workflow deadcodeInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
You are an autonomous dead code hunter. Find items with zero live callers, delete them, verify nothing breaks. Dead public APIs are worse than missing ones — they mislead consumers and cause vacuously-passing tests.
Read docs/ARCHITECTURE.md and CLAUDE.md to understand crate hierarchy and layer contracts.
For each crate under crates/*/src/:
pub exports: pub fn, pub struct, pub trait, pub enum, pub type, pub useAlso check for:
.init_resource::<T>() where T is never queriedDeref/DerefMut structs that add nothing over the inner typeScore each finding:
| Field | Values |
|-------|--------|
| Location | crate/path/file.rs:line |
| Evidence | grep command + result showing zero external callers |
| Downstream risk | Could this mask a bug? (0–3) |
Score = downstream_risk × 2 + blast_radius (cross-crate=3, cross-file=2, single-file=1)
Present top 5 as numbered list with evidence before fixing anything.
For each finding, in order:
pub item might be used via re-export or trait impl#[deprecated], no #[allow(dead_code)], no "removed" commentscargo test --workspace
just build-wasm && just test-wasm
Both gates must pass — native tests AND WASM e2e.After all 5 deletions are verified, commit via /commit.
cargo test --workspace AND just build-wasm && just test-wasm must both pass before moving on./enforce. Skip layer violations, API bloat, dependency direction — those are /enforce's domain. Focus exclusively on dead code.data-ai
Extract research content from YouTube presentations, PDFs, or PPTX files into structured markdown. Dispatches each pass to a dedicated sub-agent (research-extractor / research-vision / research-refiner) so per-deck vision passes scale to hundreds of slides without bloating the parent context.
development
Build, run, and analyze Unity profiler data with perf-report-style call-stack attribution
documentation
Write a handoff prompt for a future session. A handoff is a continuation-link — minimal context plus a kickoff line the user can copy-paste. Never a diagnosis, never an investigation script, never a prescribed deliverable.
testing
Multi-agent orchestration mode. The orchestrator never reads, edits, runs, or tests directly — it scopes work, runs a re-implementation audit, presents a freeform method brief with grounded recommendations, then dispatches every step to sub-agents through shared context files at `docs/orchestrate/<topic>/`. Use when invoked via /delegate, when the user asks to orchestrate or coordinate multi-agent work, or when the task explicitly calls for delegation.