CassioRoos
User19 protocol-driven skills for Claude Code & Codex CLI: an adversarial reviewer, its evidence-grounded analysis spine, a production-ops discipline suite, and the gauntlet that A/B-tests any skill against a no-skill control — three of these lost to loading nothing until it was run.
Categories
Indexed Skills (20)
evidence-grounding
Ground every challenge, claim, and alternative in real evidence. Search codebases for existing patterns, find real-world precedents and post-mortems, and validate claims against data. Use when making or evaluating technical claims, proposing alternatives, or challenging decisions. Triggers on "show me evidence", "prove it", "is this actually true", "back this up", "is this grounded", "evidence check", or whenever a challenge needs substantiation. Also the canonical evidence-standards reference that the analysis-spine skills link to instead of restating.
godfly
Adversarial collaborator who pushes back like a senior engineer - the default adversarial reviewer. Surfaces hidden assumptions, grounds every challenge in evidence, presents real counterpoints with proven alternatives, drives toward synthesis, and records settled claims as verdict nodes in docs/verdicts/ so an investigation is never paid for twice. Challenges code, architecture, PRs, incident fixes, product strategy, and plans. Use when user wants to "challenge this", "review this", "stress test", "poke holes", "pushback", "call my bullshit", radical candor, hard review, asks whether a claim was already settled, or types *spar, *review, *assumptions, *alternatives, *fail, *pre-mortem, or *invert. For constructing the case FOR the alternative use devils-advocate; for attack-path launch review use red-blue-review; for "what am I missing" use assumptions-check.
root-cause
Causal-chain analysis after something already happened: trace a symptom to its systemic cause using evidence-backed Five Whys and first-principles constraint challenges. Use for root cause, why did this really happen, underlying or systemic cause, and incident causal analysis once the immediate failure is understood. For a system failing right now use troubleshooting-investigator; for understanding a working system use deep-dive; for writing the postmortem document use premortem-postmortem, which uses this skill for its causal-analysis step.
competing-hypotheses
Map the full solution space and evaluate approaches against evidence. Based on the CIA's Analysis of Competing Hypotheses (ACH). Use for DECISIONS - choosing between technologies, architectures, or strategies - or when the user needs counterpoints and alternatives. Triggers on "what are the alternatives", "compare approaches", "which is better", "counterpoint", "other options", or any decision between multiple options. For ranking bug hypotheses during active debugging, use the hypothesis table in morpheus references/troubleshooting.md instead.
gauntlet
Empirically test whether a skill actually improves model output — before trusting it. Runs a controlled experiment: planted-flaw fixture, no-skill control arm, skill arm(s), optional cross-model arms via installed CLIs, blind judging with shuffled labels, and a pressure test for verdict stability. Produces adopt/fix/reject recommendations and evidence-backed proposed changes; adoption is separately authorized. Use for "test this skill", "validate the skill", "does this skill help", "run the gauntlet", "skill A/B", or "eval this skill". Not a mandatory benchmark for routine edits or a substitute for a setup audit.
toolshed
Toolshed — durable, model-agnostic working state for ONE coding task (feature, investigation) as docs/work/SLUG/ under docs/, DELETED at close. Seeds STATE + decisions/questions/evidence with grades and reproduction commands. Use for "start a toolshed", "seed the task", "/toolshed start|resume|session-end|close", or "where were we". Session end = thin STATE update. Full packet only when explicitly requested or on close/ship gates. Evidence may also live in PRs, CI, mean-qa, morpheus, monitoring runbooks — toolshed is a state shelf, not the whole environment. Pair with morpheus, handoff, safe-ops, spec-adr-builder.
assumptions-check
Build and test an explicit inventory of hidden assumptions using the Key Assumptions Check technique. Use for assumption check or what are we assuming. For general engineering review or what am I missing, use morpheus.
failure-analysis
Identify and prioritize concrete component and dependency failure modes, their consequences, containment and recovery. Use for FMEA, reliability assessment, single points of failure, or "how could this fail". General engineering verdicts belong to morpheus; active debugging to the troubleshooting reference in morpheus; past incident analysis to premortem-postmortem.
handoff
Create a compact handoff document for another Codex session or agent to continue the work. Use when the user asks for a handoff, next-agent context, continuation note, session summary, resume brief, or wants another agent to pick up without reconstructing the thread.
sentry-brief
Turn a raw alert payload into a one-screen incident brief: blast radius, suspected cause, the three checks that discriminate between causes, and the containment action. Use when paged, when triaging an alert, or when handing an incident to the next responder.
troubleshooting-investigator
Structured troubleshooting and bug investigation workflow for something failing NOW. Use when debugging failing tests, production issues, regressions, flaky behavior, errors, logs, broken builds, performance problems, or unclear symptoms. For tracing why a past failure really happened use root-cause; for understanding a working system use deep-dive; for anticipating future failures use failure-analysis.
premortem-postmortem
Run premortems before risky work and postmortems after incidents or failures. Use when the user asks for premortem, postmortem, incident review, retrospective, launch risk, "how will this fail", or "what killed this". Build the causal chain from evidence and produce the premortem/postmortem document with owners and verification.
red-blue-review
Examine concrete attack and failure paths, then design prevention, detection, containment and recovery. Use for red team, blue team, threat models, attack paths or defense plans. General engineering review belongs to morpheus; component failure inventories belong to failure-analysis.
deep-dive
Deep technical investigation for complex systems, unfamiliar codebases, architecture, data flows, and dependencies - when the system works but is not understood. Use when the user asks for a deep dive, investigate deeply, map the system, understand how it works, break down complexity, or produce an evidence-backed technical brief. If something is failing right now use the troubleshooting reference in morpheus; for a past incident timeline and contributing factors use premortem-postmortem.
spec-adr-builder
Generate or review technical specs and Architecture Decision Records. Use when the user asks for specs, ADRs, RFCs, design docs, decision records, technical requirements, non-goals, alternatives, rollout, or rollback documents. For user-research requirements elicitation use user-research.
deployment-monitor
Monitor deployments with read-only evidence gathering, anomaly detection, cadence summaries, and local alerts. Use when watching a deploy, validating staging or production after release, comparing Datadog/database/runtime evidence, investigating new errors or skipped work, or linking post-deploy anomalies back to GitHub code, commits, PRs, or issues.
morpheus
Second opinion and final verdict for engineering decisions. Challenges a plan, design, PR, claim, fix, or bug theory with cited evidence, steelmans the user's position first, asks when an assumption would flip the verdict, labels how each claim is known, and always lands on a verdict or a recommended option with tradeoffs. Use when the user says "challenge this", "review this", "poke holes", "stress test", "call my bullshit", "second opinion", "should I", "give me options", "what do you think", or when the orchestrator holds a non-trivial recommendation and wants pushback before committing. Includes focused troubleshooting and research guidance; deployment watches are explicit-only. Not for formatting or trivia.
devils-advocate
Argue the opposing position: construct the strongest possible case for the alternative or against a pending decision. Use for make the case for NOT doing this, argue the other side, steelman the alternative, what would critics say. Produces an opposing-counsel brief, not a code review. For evidence-based critique of your own work, plans, or PRs use morpheus; for attack-path and launch-readiness review use red-blue-review; for imagining failure before starting use premortem-postmortem.
safe-ops
Safely perform authorized remote, destructive or security-relevant operations, including Git push/merge, releases, infrastructure and data mutations, credential changes and bulk cleanup. Not required for routine read-only inspection or ordinary scoped repository edits.
mean-qa
Adversarial QA campaigns with persistent .proof evidence and re-runnable cases. Use to QA a feature, test a browser flow, verify a change or staging build, design a test campaign, or prepare a PR verification section. Local evidence is mandatory; publishing is a separate authorized action.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.