Stop shipping features your AI agent invented. A Claude Code plugin that wraps your agentic-coding loop with three filters: Devil's Advocate before every plan, mandatory 10% scope kill, drift checks every 5 features. Built on the Ralph Loop.
Agent
Use during PM mode BUILD with FRESH context per feature. Implements the minimum code to pass the tests written by test-writer. Max 3 resets per feature. Reads architecture.md (relevant module only) + progress.md top sections + contracts.md if applicable. Flags REWORK / BREAKINGCHANGE / TECHDEBT. Qualityscore is…
Agent
Use to falsify any plan, pitch, PRD, thesis, or release before commitment. Attacks every hypothesis, runs a pre-mortem (≥5 scenarios), audits evidence for cherry-picking and source-diversity skew, stress-tests defensibility (replicability, erosion, network effects, Thiel's 7), separates existential vs non-existential…
Agent
Use during PM mode BUILD every 5 completed features, or anytime the user wants a drift audit. Audits the last 5 features against IDENTITY.md (WHO/JOB/NEVER), Architecture Constraints, contracts (if contractmode != "none"), and rework patterns. Returns CONTINUE / DISCUSS / PIVOT / CONTRACTHALT and flags STABILIZE if…
Agent
Use after a SHAPE plan is approved in PM mode to decompose the plan into buildable features for the Ralph Loop. Additive mode for existing products (extends features.json without re-implementing). Every feature gets signalid + requestedby + autonomy tier + dependson. Sized for a single context window (3-7 acceptance…
Agent
Use during Founder mode INTERROGATE phase — generates Mom-Test + JTBD interview protocols (10–15 questions traced to hypotheses), analyzes raw transcripts, extracts JTBD four-forces (Push/Pull/Anxiety/Habit), tags evidence by weight (behavioral=3, mixed=2, stated-intent=1), and flags false positives (innovation…
Agent
Use when scouting communities for product signals during Founder mode SCOUT phase, or when the user wants to mine Reddit/G2/HN/ProductHunt/Crunchbase for pain, willingness-to-pay, competitor moves, or contradictory evidence on a hypothesis. Returns a structured SCOUT report with verbatim quotes, source URLs, and new…
Agent
Use to manage a full Forge from start to finish, in either Founder mode (SCOUT → INTERROGATE → SYNTHESIZE → TEST) or PM mode (HARVEST → SHAPE → BUILD → SHIP). Auto-detects mode from state files. Enforces phase order, mandatory Devil's Advocate after every INTERROGATE, mandatory drift check every 5 features, mandatory…
Agent
Use during PM mode BUILD before the builder runs. Writes 3–7 acceptance tests per feature in the project's test framework. For API features when contractmode="full", contract-compliance tests come FIRST. For removal features, includes "is gone" + "adjacent features unaffected" tests. Tests are the spec — they must…
Agent
Use during Founder mode SYNTHESIZE phase. Integrates new evidence into THESIS.md, scores per-section confidence on 1–5 rubric with hard caps (single source → cap 3, all stated-intent → cap 3, 5 requires behavioral + commitments), audits claims for UNSUPPORTED / WEAK / CONTESTED, tracks trajectory across cycles…