featbit-release-decision-agent

Makes product or AI changes reversible before they are visible, and defines who sees the change, how much traffic, and how expansion will be controlled. Activate when triggered by CF-03 or CF-04 from the release-decision framework, or when user says "feature flag", "should I ship this", "rollout strategy", "gradual rollout", "canary", "5% of users", "start exposing", "who sees this first", "create a flag", "add flag to code", or asks how to hand flag requirements to another team. Default to producing a clear implementation handoff when the current user does not own code or flag operations; FeatBit CLI and Web UI references are optional operator adapters.

Evaluates collected data to determine if evidence is sufficient to decide, then frames the outcome as CONTINUE, PAUSE, ROLLBACK CANDIDATE, or INCONCLUSIVE. Activate when triggered by CF-06 or CF-07 from the release-decision framework, or when user says "analyze results", "should I ship this", "continue or rollback", "is this significant", "what do the results say", "has it been long enough". Do not use when data collection has not started.

Creates and manages experiments as database-backed records — the unified replacement for an online A/B test dashboard. Activate when user is ready to start collecting data for a hypothesis, wants to set up an experiment, needs to pull or organize experiment data, wants to run analysis calculations, or asks "how do I track this experiment", "set up the experiment", "run the analysis", "pull results", "compute the stats". Sits between measurement-design (instrumentation confirmed) and evidence-analysis (results interpreted). Do not use before a hypothesis and primary metric exist.

FeatBit release decision philosophy and control framework. Activate when the user is anywhere in the loop from intent to implementation to feedback to next iteration and needs the right decision lens, not a fixed workflow. Triggers — "release decision", "what should we build", "feature flag", "A/B test", "experiment design", "how do I measure this", "should I ship this", "rollout strategy", "analyze results", "continue or rollback", "how do I know it worked", "what did we learn", "next iteration", "optimize page", "increase adoption".

Converts a clear business goal into a falsifiable hypothesis before implementation begins. Activate when triggered by CF-02 from the release-decision framework, or when a goal exists but there is no explicit causal claim linking a change to an expected outcome. Triggers — "write a hypothesis", "what do we expect", "what should we test", "we think this will work because". Do not use when the hypothesis is already sharp and falsifiable.

Extracts the real business outcome when the user has a vague direction or jumps to tactics before the goal is clear. Activate when triggered by CF-01 from the release-decision framework, or when user says "I want to improve X", "we should add Y", "increase adoption", "make it better", or describes a tactic without stating a goal. Do not use when the goal is already measurable and specific.

Captures structured learnings at the end of an experiment or release cycle so the next iteration starts from evidence rather than memory. Activate when triggered by CF-08 from the release-decision framework, or when user says "what did we learn", "close this experiment", "we're done with this cycle", "next iteration", "this experiment is over", "capture learning". Activate immediately after a decision is made in evidence-analysis.

Defines the primary success metric, guardrails, and event instrumentation for a hypothesis before exposure begins. Activate when triggered by CF-05 from the release-decision framework, or when user asks "how do I measure this", "what metrics should I use", "what should I track", "how do I instrument this", "event design", "what is the primary metric". Do not use when metrics are already well-defined and instrumentation is complete.

Sync release-decision experiment state to the web database. Provides the `sync.ts` CLI script that all other skills call to persist state changes (update fields, advance stage, log activities, manage experiment runs). Activate whenever a skill needs to write experiment state to the web DB. Triggers — "sync to web DB", "update experiment state", "push state", "set stage", "add activity", "create run", "start run", "analyze run", "decide run", "archive run", "record decision", "save learning", "get experiment".