mtschoen

Use after you've finished and verified a change (smoke test passed, or smoke-test concluded N/A for a prose-only change) and have no further edits planned - before declaring work done, committing, pushing to main, or opening a PR. Checks whether the change…

Use when about to commit a draft solution that suppresses a problem (silent fallbacks like || true, [ExcludeFromCodeCoverage] on throw/stub code, lowered thresholds), pulls state from somewhere illegitimate (binaries from an unrelated app's install dir, prod…

Use when you finish implementing a feature, fixing a bug, refactoring, or changing configuration - any change that affects runtime behavior - before saying 'done', 'complete', 'finished', 'all set', 'ready to commit', or any variation, even if you're…

Use when the user says /wrap, "wrap up", "close out this session", "finish the session", or otherwise signals intentional session end. Also when the user asks to update memory, save learnings, or commit everything before exit. Runs only when the user…

Use when the user asks for a maintenance pass, cleanup, end-of-session tidy-up, or health check on a repo. Triggers: 'clean up this project', 'run maintenance', 'end of session cleanup', fleet sweeps that follow project-tracker.find_stale_maintenance.…

Use when the user asks the agent to reconcile a project's PLAN.md/TODO.md against its recent git history - e.g. 'what's actually done in project-tracker?', 'audit completed work for this project', 'any tasks I should check off?', 'reconcile the plan with what…

Use when: user mentions coverage, lint, linter, static analysis, code quality checks, 100% coverage, coverage gate, lint baseline, test report, TEST-REPORT.md, or 'is the build clean'; BEFORE declaring work done, summarizing what you built, or saying 'all…

Use for ANY debugger or memory-forensics work in C#/C++ (and similar native/managed runtimes) - not just bug hunts: verifying runtime behavior, inspecting a healthy process, heap/GC analysis (SOS, dumpheap, gcroot, "who holds a reference to this object"), and…

Use when changing code shape without changing behavior - renaming or deleting symbols, changing signatures, extracting methods or modules, moving code, or converting a data shape - especially when the change must touch many call sites. Triggers: about to…

Use when the agent observes slow test runs in this session (multi-minute wall clock, repeated re-runs blocking iteration, tests dominating the inner loop, thousands of subprocess spawns, coverage/instrumentation overhead) OR when adding tests likely to slow…

Use on EVERY user request that proposes work, changes direction, or makes a technical claim - code changes, new features, refactors, pivots, "while you're in there" asks, choice of approach, statements about how the code works. The agent often has code and…

Use when about to do extended Read/Grep/thinking on a question whose answer is observable behavior of an external system, library, framework, or runtime - prefer learning from running code in a scratch dir over inferring from docs. Suppressed when the…

Use before designing, planning, writing custom code, refactoring/extracting code, choosing a tool/library/protocol, or debugging an unfamiliar error. Steers toward running a WebSearch (or Read/Grep for codebase-only questions) BEFORE acting on a decision,…

Use when the user asks for retrospective Claude Code cost or active-time analysis over a date range, including spend, top sessions, cache discipline, subscription leverage, time spent waiting, or time spent on slash commands such as /wrap.

Use when delegating work to a different machine because of hardware, OS-specific tooling, locally-installed services, or anything else that physically can't be done on the orchestrator's host. Triggers include "I need to run this on the linux box," "the GPU…

Use when diverting work from the metered primary harness to other installed agent CLIs to conserve its quota or to reach models it does not offer. Triggers include "run this on a cheaper harness", "save my Claude tokens", "use another subscription for this…

Use before making changes in any project or repository, especially when a session reaches outside its starting working directory. Checks for a cooperative project lock, acquires one before writes, chooses wait versus worktree when another agent holds it,…

Use when orchestrating 2+ implementer agents in parallel against isolated branches or worktrees and you reach a merge point - branches returning with green reports, deciding what to merge, declaring a round done, or moving to the next parallel phase. Fires on…

Use when dispatching subagents across multiple LOCAL PROJECT REPOSITORIES - feature implementation, maintenance sweeps, or fleet-wide investigation across the user's project directory. Triggers: 'spawn agents to fix X across all my projects', 'run a…

Use during any non-trivial turn (multi-file edits, multi-step research, planning + implementation, dispatching subagents, or anything you'd ballpark at over 2 minutes wall-clock). Periodically emits a progress-beacon JSON block in the assistant message text…

Use when the user wants to change the agent-statusline status line live - hide or show cost figures, switch compact mode, set or derive the burn-rate target, set a daily budget, toggle verbose pace, or turn the progress-beacon ETA on/off - without restarting…

Use when auditing one or more machines before travel, migration, shutdown, repair, or cleanup for dirty tracked files, unpushed branch commits, or git stashes.

Use when about to commit a draft solution that suppresses a problem (silent fallbacks like || true, [ExcludeFromCodeCoverage] on throw/stub code, lowered thresholds), pulls state from somewhere illegitimate (binaries from an unrelated app's install dir, prod…

Use when the user asks for a maintenance pass, cleanup, end-of-session tidy-up, or health check on a repo. Triggers: 'clean up this project', 'run maintenance', 'end of session cleanup', fleet sweeps that follow project-tracker.find_stale_maintenance.…

Use when the user asks to clean up, audit, prune, consolidate, or do maintenance on their agent memory / notes corpus - triggers include "memory cleanup", "clean up my notes", "audit the notes corpus", "prune stale memory", "consolidate memory notes", "run…

Use when about to act on a topic the notes corpus may already cover, or when the user asks what memory exists about something - searches the shared notes corpus for relevant facts, with an optional per-project narrowing flag. Triggers include "check memory",…

Use when a brainstorm has fixed a project's tech stack and the user wants to start real work - "promote this to a real project", "scaffold this", "set up the bells and whistles", "make this a real repo", "wire up the linters and hooks". Also when adopting an…

Use when the user asks the agent to pick or suggest a task to work on across their tracked projects. Triggers include "find me a task", "what should I work on", "pick me something fun/quick/meaty", "something fresh to dig into", "help me find something to…

Use when the user wants to externalize a brainstorming conversation into a new project folder. Triggers include "capture this as a project", "let's start a folder for this", "jot this down as an idea", "put this somewhere I won't forget". Creates a folder via…