Evidence-bound investigation producing a falsifiable architectural mental model of a code concept, pattern, term, or component in large legacy codebases. Use WHENEVER user asks WHAT/HOW/WHERE about code — /explain, "explain X", "what is X", "how does X work", "trace this concept across layers", "preflight before bug-fix", "build mental model", "architectural overview", "layer-by-layer explanation", "F/B/D-XXX pattern explanation", or pastes a domain term or component name asking what it does or how it fits. Four modes — --teach (default educational), --brief (≤5pp preflight), --for-bug-fix (≤3pp "Where to dig" 1-pager feeding /bug.why), --deep (full + alternative-design). Falsification-first, anti-hallucination — Step-Back + Architectural Hypothesis BEFORE any code, AST-aware retrieval, hybrid grep+graph subagent fan-out, Cross-Layer Contract Violation Detector, Layer Activation Matrix, ≥1 ◆◆◆ per key claim with Mutation Counterfactual, CoVe Gate 2 factored verification, Citation-Grounded re-read, Mental-Real
Pre-implementation blueprint for AI coding agents on legacy codebases. Use WHENEVER user asks HOW to implement, plan, or design — /howto.implement, /blueprint, "implementation plan / analysis / strategy / approach", "how to implement / add / build / refactor", pastes Jira / user story / RCA / SPEC asking "plan this", or mentions Tree of Thoughts, spec-driven development, Design-by-Contract. Falsification-first, anti-hallucination — Approach
Evidence-bound transformation of vague human requirements (Jira ticket, Confluence page, free text, file path) into AI-ready specifications for coding agents. Use WHENEVER user asks to /distill, "analyze requirements", "clarify task", "vague requirements", "ambiguous AC", "task vs bug", "EARS", "Given-When-Then", "GWT", "AI-ready spec", "prepare spec for implementation", "Open Questions", "INVEST", "MoSCoW", "user story analysis", "acceptance criteria", "steps to reproduce", "expected vs actual", or pastes Jira-key / Confluence URL / free-text requirement asking "make this implementable / specifiable / unambiguous". Three modes — --brief (≤5pp preflight before /howto-implement), --standard (default Tier 2 full SPEC with EARS+GWT+traceability+glossary), --deep (Tier 3 security boundary / public API / migration / financial; multi-agent k=3 vote on ambiguity). Falsification-first, anti-hallucination — Killer Requirement FAIL-FAST, Goal Anchor Step-Back, Anti-Anchoring KNOWN/OBSERVED/UNKNOWN BEFORE normalization,
Root cause analysis for software bugs in large legacy codebases. Use this skill WHENEVER the user asks WHY a bug happens, asks for RCA, mentions root cause analysis, says /bug.why, /bug.investigate, asks "why is this failing", "find the root cause", "trace this exception", "investigate this bug", "post-incident review", or pastes a stack trace, error log, exception, NullReferenceException, SqlException, InvalidOperationException, or asks 5 Whys / Kepner-Tregoe / IS-IS-NOT / fault localization / hypothesis-driven debugging / falsification / counterfactual debugging / fault tree / blast radius / evidence weighting / Bayesian debugging / error propagation. Falsification-first, evidence-bound, anti-hallucination by design — generates Killer + H-Data + 5–7 hypotheses with 6C+D coverage, runs discriminating experiments before evidence collection, applies asymmetric Bayesian updates, gates on CoVe re-read + Mutation Counterfactuals + Backward Dataflow Spot Check, declares INCONCLUSIVE rather than fabricate. Investig
Surgical code-execution skill for AI coding agents on legacy codebases. Use WHENEVER the user asks to IMPLEMENT, CODE, BUILD, APPLY, or FIX from a contract source — IMP / Plan + TASK-NN, RCA report, freeform task, or Jira ticket. Triggers on /implement, /bug.fix, /code, /apply, /fix, "implement this", "code this", "build this", "apply this fix", "fix this bug", "execute the plan", "implement the blueprint", paths like `IMP-*.md`, `PLAN-*.md`, `RCA-*.md` followed by "implement"/"fix"/"apply", or pasted Jira ticket asking "implement / fix / build". Two MODES — IMPLEMENT (from IMP / Plan + TASK-NN / freeform) and FIX (from RCA Section 7.1 PRIMARY fix). Surgical-changes-first, verifier-guided, falsification-first, anti-hallucination — Read-Before-Edit + 3-line anchored edits, Stale-Context check, YAGNI master checklist, Diff-Budget hard cap (1.5× planned), Semantic Diff Guard (forbidden orthogonal damage), Verifier-Guided Self-Refinement Loop (run → on FAIL revise → rerun, max 3), Mutation Challenge on the fix, a
Hypothesis-driven task-correctness verification for AI coding agents on legacy codebases. Use WHENEVER user asks to verify, audit, or pre-flight a task artifact (IMP, RCA, Plan, Implementation, Fix, Jira ticket) before submission — /verify, /verify-task, /gate2, "verify this", "verify the RCA", "verify the implementation", "is this fix correct", "double-check my plan", "audit before merge", "pre-flight before PR", "Gate 2 / Pre-Send check", "find what's wrong with my IMP", or pastes RCA-/IMP-/PLAN-/FIX-*.md asking "verify". Falsification-first, evidence-bound, anti-hallucination — Architecture Pre-Read MANDATORY for multi-layer artifacts; Initial Intuition + KNOWN/OBSERVED/UNKNOWN anti-anchoring table BEFORE reading artifact reasoning; Killer Hypothesis FAIL-FAST; CoVe factored independent re-read; Stale-vs-Hallucinated decision (file shifted vs never existed); Pre-Claim Symbol Existence Check (Grep/Bash/WebSearch every external method/class/flag/pattern/CVE the artifact claims); Tier-budgeted hypotheses (5–3
White-box code-path tracing in two modes. REAL — empirical verification of completed code via CFG/MPC traversal, Counterfactual + optional Live Mutation Testing, PBT hook, Tandem-FL on FAIL, Killer-Path-First, FAIL-FAST, persistent index-prefixed JSONL trace with resume + confidence decay, Reviewer Agent re-rank, Reflexion lessons, structured fix_proposal + repro_snippet. VIRTUAL — pre-implementation falsification of RCA reports / Implementation Blueprints / Plans / freeform ideas via Planner-Simulator, POPPER falsification-first, LTL plan consistency, cross-artifact drift detection, defended against Mental-Reality Gap. Use WHEN user says /trace, /trace.real, /trace.virtual, /verify-plan, "verify this code", "trace these paths", "check coverage", "verify this plan", "falsify my RCA", "is this blueprint sound", "will this implementation work", "will my fix work", "white-box tracing", "mutation testing", "CFG analysis", "MPC", "SBFL", "Tandem-FL", "fault localization". Investigation only — never modifies source