Explores 3-5 distinct approaches before committing to one. Use when: Design decisions: "design", "architect", "plan", "how should we", "what's the best way" Comparing options: "options", "alternatives", "trade-offs", "pros and cons", "compare approaches" Decision making: "help me decide", "how should I implement"…
Enforces verification over recall and prevents scope creep. Use when: Acting on a claim from docs, project memory (AGENTS.md/CLAUDE.md), or a ticket without checking the code: "the migration is complete", "X is handled by Y", "the flag is enabled" Guessing a function signature or import path from memory Modifying…
Prevents code duplication by requiring inventory check before creation. Use when: Creating new code: "create new", "add a service", "build a helper", "write a utility", "new class" Adding infrastructure: "implement middleware", "add validator", "create repository", "new module" Building shared code: "shared…
Guides users through PM workflow phases and command sequence. Use when: Asking for guidance: "what command", "which phase", "where do I start", "what's next", "workflow" Starting work: "new project", "new session", "help me plan", "kick off", "get started" Learning the workflow: "how do I use this", "what are the…
Requires evidence before any completion claim. Use when: Checking status: "is it done", "is this done", "are we finished", "ready for review", "create PR", "commit" Claiming completion: "ready to close", "mark as complete", "verify my work", "did I miss anything" Shipping code: "ship it", "LGTM", "merge it", "push…
Use when running /close-epic, when about to file the epic closure comment, when an operator asks "what did we consider but not pursue in this epic", or when about to summarize an epic's surfaced-but-deferred work. Also use when about to grep sub-ticket comments for closure-log entries by hand, when about to skip a ###…
Use when Codex review returns findings, when processing codexreviewandfix or codexreview results, when collecting a codex-delegate job with codexstatus/codexresult, when cross-model review report contains P0-P3 items, or when about to post a Codex review report to Linear. Also use when about to file a Codex finding as…
Use when about to defer in-scope work or file a follow-up ticket for something observed during execution — saying "I'll defer this", "follow-up ticket", "out of scope", "future work", "handle separately", writing a Deferred Items entry, classifying AC-DEFERRED, or advancing a phase with known unfinished in-scope work.…
Use when user asks "is the workflow actually working", "are we deferring too much", "what's our profile mix", "did the impact bar help", "how many follow-ups did this epic file", "what's the codex auto-fix rate", or any other meta-question about pm-vibecode-ops execution patterns. Also use when about to count Linear…
Use when any workflow phase completes for a ticket — adaptation, implementation, testing, documentation, code review, codex review, or security review. Use when processing agent results, when about to advance to the next phase, when a ticket status is being updated, or when an agent has returned a structured report.…
Use when a test fails, a build errors, a runtime exception appears, or behavior is unexpected. Also use when about to "just try" a fix without stating a hypothesis, when the second or third fix attempt has failed, when tempted to change multiple things at once, or when saying "it's broken", "not working", "failing"…