m0n0x41d/haft

Decision Verification checks whether a recorded DecisionRecord still holds by running baseline vs measure evidence loops and drift detection. Use it when a shipped decision needs a post-implementation reality-check before further work relies on it.

H Explore generates 3-5 genuinely distinct candidate approaches for a current question, keeping the weakest link of each visible. Use it when alternatives are needed or a favored approach is prematurely closing the search.

H Status reports read-only project memory and coverage from a current code index, including active problems, decisions, notes, evidence freshness, drift, and file-link gaps. Use it for project status, session resumption, or answering what is decision-linked, uncovered, or needs attention.

H Verify compares recorded DecisionRecords or claims against current code, tests, measurements, or incidents to surface weakened, refuted, stale, or drifted claims. Use it when design-time claims need runtime evidence before relying on them.

H-Commission creates a WorkCommission grant from an active DecisionRecord only when manually invoked by an operator. It verifies source decision status and freshness, derives execution scope, snapshots the autonomy envelope, and stops before any execution.

H-Commission creates bounded execution authority from an explicit operator grant and never auto-triggers or infers approval from prompt text or tool schemas. Use it when an agent needs a manual human gate before executing scoped actions.

H-Compare compares two or more existing candidates under an explicit characteristic space, parity basis, and predeclared selection policy, returning trade-offs and a non-dominated set. Use it when you need rigorous multi-candidate comparison rather than a binding choice.

H-Decide routes one direct, unambiguous operator request to bind a bounded choice as a DecisionRecord. Use when the operator asks to decide, bind, or supersede a choice; h-frame, h-explore, and h-compare are independent capabilities, not mandatory phases.

H-Decide routes one direct, unambiguous operator request into a bounded binding choice using the haft CLI or MCP. Use it when an operator asks to bind, decide, or supersede an exact choice and provenance must be recorded honestly.

H-Diagnose identifies the most likely cause of a concrete failure by stabilizing the signal and testing multiple rival explanations in parallel read-only probes. Use it when symptoms are observable but the root cause is unclear, and preserve losing hypotheses for fallback or replay.

H-Diagnose stabilizes a concrete failure, generates distinct rival explanations, tests discriminating observations read-only in parallel, and ranks them by evidence while retaining losing hypotheses for later use.

H-Explore generates 3–5 genuinely distinct candidate approaches for a current question, with the weakest link of each visible. Use it to keep alternatives live, inspect direct pattern bodies, and avoid premature convergence when exploring uncertain design or reasoning paths.

H-Frame shapes an engineering problem before a solution or project phase is assumed, separating observation from cause and proposed fix. Use it when a redesign, refactor, or proposal arrives before affected object, constraints, and acceptance basis are clear.

H-Frame shapes an under-articulated engineering problem without assuming a solution or forcing a project phase. It retrieves current first-principles source and creates a ProblemCard only on explicit save intent or a concrete receiving use.

H-Note persists a fact, observation, caveat, or small non-binding rationale in Haft project memory when the operator explicitly asks or current work supplies a concrete receiving use. Use it for 'remember this', 'запиши', or 'note for later' requests.

H-Note saves lightweight facts, observations, caveats, or small rationales to Haft project memory when the operator explicitly asks to remember something. Use it to persist non-binding context that later tasks can reference, not for ordinary reasoning.

H-Onboard bootstraps Haft through one readable onboarding surface, prepares initial profile reviews or bounded scope relation changes, and orients specification carriers. Use for first-time setup, incomplete setup, or profile-underdetermined spec recovery.

H-Reason is a source-first umbrella for FPF-aware reasoning in a Haft project. Use when the operator asks to think through an ambiguous engineering, management, architecture, specification, or project question without naming a narrower Haft capability.

H-Reason provides source-first FPF-aware reasoning for ambiguous engineering, management, architecture, specification, or project questions in a Haft project. Use it when no narrower Haft capability is current; ordinary reasoning stays conversational and binding actions remain manual.

H-Spec manages typed specification lifecycle and source-currentness repair: inspect current SpecSections, draft or clarify carriers, classify FPF semantic fanout, record operator-requested spec changes, and cross explicit approve/rebaseline/reopen gates only with human authorization. Use for spec status, updates, stale sections, newer source revisions, or semantic changes across carriers.

H-Spec manages Haft's typed spec lifecycle, grounding drafts in repository evidence and current carriers, validating authored drafts, and repairing semantic fanout and L/A/D/E. Use it when authoring, tracing, or approving specs in Haft with explicit human gates.

Haft Compare compares two or more existing candidates under an explicit characteristic space, parity basis, and predeclared selection policy, returning constraints, trade-offs, and a non-dominated set rather than a single score. Use it when you need to evaluate options for an AI agent without scalarizing into one hidden ranking.

Haft Onboard bootstraps Haft through the readable task-level onboarding surface, prepares a non-binding project-profile review when needed, and orients only applicable typed spec carriers. Use it to keep profile apply and lifecycle gates human while setup is ready immediately after haft init.

Haft Status provides a read-only cockpit of a Haft project's active problems, decisions, notes, evidence freshness, drift, commissions, spec lifecycle, module coverage, and exact file-link gaps from the current code index. Use it for project status, session resumption, finding decision-linked items, or spotting what needs attention.