adr
Architecture decision records: durable decisions with their date, context, and consequences — supersede, never rewrite.
The performance discipline: measure before touching, benchmark what matters, prove regressions and fixes with numbers.
> /plugin marketplace add azrtydxb/procoder > /plugin install procoder@procoder
How it fires
How this command gets triggered: by you, by Claude, or both.
/perfContext preview
What this command does when you run it.
The performance discipline: measure before touching, benchmark what matters, prove regressions and fixes with numbers.
description: "The performance discipline: measure before touching, benchmark what matters, prove regressions and fixes with numbers."
The user invoked /procoder:perf with arguments: $ARGUMENTS
The launcher is: "${CLAUDE_PLUGIN_ROOT}/hooks/launcher.sh"
Performance work without numbers is storytelling. The discipline:
1. **Baseline first.** Run `launcher.sh bench` — it runs the Go benchmarks and compares against `.procoder/bench/baseline.txt`. No baseline yet? `launcher.sh bench --save` records one BEFORE you touch anything. No benchmarks yet? Write one for the hot path the change touches — that is the first task, not an afterthought. 2. **Change, then measure again.** `launcher.sh bench` marks any regression beyond `[bench] threshold` (default 10%) and exits 1. Improvements are numbers too — quote the delta, not adjectives. 3. **Save deliberately.** A new baseline (`--save`) is a decision that the current numbers are the new normal — take it after a reviewed improvement, never to silence a regression. 4. Results are single-run and machine-local; treat small deltas as noise and cross-platform comparisons with the warning the tool prints. For deeper hunts, profile (pprof) before optimizing.
Senior-developer discipline for AI coding agents. A commit gate that counts unchecked as failing, quality controllers that refuse to call unfinished work done, and a lessons loop that closes each escaped bug's class. One Go binary, no runtime deps, works with 20+ agents.
Architecture decision records: durable decisions with their date, context, and consequences — supersede, never rewrite.
Onboard an existing codebase: every domain's checks over the whole tree, then a triaged plan to bring it in line.
The project layer: milestones, epics, and user stories with refusing controllers — spec-seeded, sprint-ready.
Run the formatting gate over the changed files, as a commit or CI would.
CI/CD hygiene and run health: pinned actions, timeouts, concurrency, tests — plus the latest runs via gh.
Copilot's auto-review escapes: find them since the last look, sanitise them, and turn each one into a lesson.