wabblespec

Standalone adversarial analysis module. Generates the strongest honest case against an artifact or decision. Invokable directly by any caller — not only Reviewer. Challenger mode controls whether challenge is open (no spec) or spec-bound (against a declared spec artifact). Do NOT invoke to issue a verdict or score — that is Grader's job. Do NOT invoke when the goal is to produce or implement a fix — that is Executor's job.

Root cause investigation using structured analysis methods. 5-whys, fishbone, fault-tree. Knowledge-graph-leveraged when Nexus is available. On-demand.

API lifecycle management. Version, deprecate, audit contracts, enforce backward compatibility. Four modes: --version, --deprecate, --contract, --audit.

Gateway routing engine. Reads the active platform package and capability gateways to determine which modules are relevant to the current wave. Assembles multi-module context using Economy placement rules. Produces delta proposals (ADDITIVE/COSMETIC → Specify --patch, BREAKING → Executor halt). Writes only to product space — never touches .wabblespec/ (I11).

Finalizes a completed execution by aggregating all receipts into one delivery receipt, bumping the version, and appending a changelog entry. A task is not done until Archive runs.

Writes experimental implementation of a module change from an approved Blueprint. For AUGMENT type: completes a modified copy of the existing module. For NEW type: completes Factory stubs. All output goes to .wabblespec/state/experiments/augments/. Never touches production module space.

Full lifecycle meta-orchestrator. Exclusively owns .wabblespec/meta.md. Scale-adaptive autonomy (L0–L4) from Decompose complexity score. Manages phase transitions, triggers Dream post-wave, schedules Evolution. Activates for multi-stage runs or /autopilot command.

Autonomously optimize a measurable metric by iterating one atomic git-committed change at a time, reverting on regression, and surfacing findings when stuck.

Quality gate before any module promotion. Runs experimental implementation against fixture set, computes held-out metric, compares against blueprint threshold. AUGMENT type requires >= 80% parity. NEW type requires >= 60% improvement over baseline. Failure halts promotion and appends to tracker.json.

Converts a Synth candidate into a formal promotion proposal. Names before/after behavior, defines benchmark gates, lists affected targets. Requires Attestation (human sign-off) before Augment or Factory may proceed.

Divergent idea generation before convergent planning. Produces a ranked option set for downstream evaluation. Invoked on-demand before Specify or Propose when the solution space is genuinely open.

Generates C4 architecture diagrams at four levels (Context, Container, Component, Code) by analyzing the codebase and producing Mermaid output. Use when a user asks for an architecture diagram, system map, C4 model, or structural overview of how the system is organized.

Converts git commit history into user-facing release notes. Conventional commit parsing, user-language translation, structured output.

Applies targeted surface-level cleanup to code. Removes dead code, normalizes formatting, renames identifiers to match spec conventions, removes deprecated patterns flagged by Specify. Scope always declared explicitly — no unbounded clean. BREAKING changes halt and route to Executor.

Stage, split, and write conventional git commits. Enforces type(scope): description format, 72-char subject line, why-not-what body, and coherent atomic units.

UI micro-text: labels, error messages, tooltips, empty states, security warnings, onboarding prompts. Concise, action-oriented, context-aware.

Breaks a locked task card into an ordered wave plan. Each wave is independently verifiable with a declared checkpoint and rollback target.

Dependency health audit. Scans for vulnerabilities, outdated packages, license issues, and supply chain risk. Produces SBOM. On-demand or pre-deployment gate.

Structured debugging protocol for hard bugs and performance regressions encountered during product-space development. Use when a bug is confirmed but the cause is unknown, a performance regression appeared, or a prior fix attempt failed.

Documentation generation for completed tasks. Produces README updates, API docs, and inline comments from verified receipts and wave artifacts. Activates after Archive receipt confirms task complete. Does not run during execution waves. For reference-type outputs, enforces a cross-references section before writing the document receipt.

EMA confidence decay and gap detection. Reads all curated drawer JSON files, applies deterministic decay, writes gap-map.md and staleness-map.md. Zero LLM involvement — script only. Triggered by session-end hook or manually.

Token density enforcement. Ensures WabbleSpec output stays within declared token budgets. Currently covered by invariant I12 (no redundant output). Separate module not yet justified — activate if I12 enforcement proves insufficient.

Coordinates multiple runtime lanes for tasks no single lane covers adequately. Triggered exclusively by ModelRouter. Three modes: Sequential (B needs A output), Independent (parallel, independent artifacts), Cross-check (both run same task, Grader compares). Cross-check required for Attestation and Audit modes. Writes combined receipt naming all lanes used.

Temporal entity relationship graph from drawer data. Extracts typed entities, builds co-occurrence graph with min threshold, assigns confidence tiers, tracks valid_from/valid_to temporal validity. Zero LLM involvement — script only.

Wave execution engine. Reads the locked wave plan, runs Guard before each wave, saves checkpoints, invokes Verifier after each wave, routes errors by type, and writes receipts throughout. On module-build tasks, verifies tests/acceptance.md exists before writing the final execution receipt.

Graph-first codebase discovery. Traverses from high-value nodes (entry points, config files, test files, dependency manifests) to produce project-map.md. Writes findings to Memory as FRESH drawers. Notifies EntityGraph for entity extraction. Per-target traversal order follows platform conventions from the active L3 package.

Scaffolds a net-new module from an approved Blueprint. Writes stub files (SKILL.md, skill-rules.json, receipt schema, tests/acceptance.md) to .wabblespec/state/experiments/augments/. Never writes to production module space.

Feature flag lifecycle management. Create, roll out, audit, and retire flags. Maintains flag manifest. Rollout gated on declared conditions.

The only L8 module that writes to production module space. Promotes a benchmarked experiment to modules/{layer}/{module}/. Gates: benchmark-passed receipt + Blueprint approval + no open contradictions in tracker.json. Self-modification always requires Attestation. After promotion: module receipt updated, experiment archived, downstream specs marked NEEDS_REVERIFICATION.

Controlled deletion of memory drawers. Writes Provenance deletion record before any deletion. Handles single, bulk EXPIRED, and compliance deletion types. Notifies EntityGraph after deletion. Never deletes without a Provenance record.

Aesthetic capability gateway. Visual design standards for any project with a visual surface — brand identity, color system, typography, motion, and design token policy. Applies to Web, Mobile, Desktop, Game, Extension/Plugin targets. Non-visual targets (API/Service, CLI, IoT, Library, Data/Pipeline) do not activate unless a visual component is declared in scope.

AI capability gateway. Cross-cutting standards for any project embedding LLM features — prompt engineering, chain design, agent architecture, evaluation, safety, and model governance. Activates on AI/Agent build target always; activates on any other target when Explore detects an LLM SDK dependency.

Design capability gateway. UX and interaction standards for any project with a visual surface — user mental models, information architecture, interaction design, design system governance, and accessibility baseline (WCAG 2.1 AA). Applies to Web, Mobile, Desktop, Game, Extension/Plugin targets. Distinct from Aesthetic gateway: Aesthetic owns the visual language; Design owns how things work and how users move through them.

Document output gateway. Routes any request to produce a Word document, PDF, Excel workbook, or PowerPoint presentation to the correct format handler with cross-cutting quality rules. Activates when archive, document, writer, report, present, visualize, or specify requests a binary document output. Summonable by any skill that needs to produce a stakeholder-ready deliverable.

Document output gateway. Routes any request to produce a Word document, PDF, Excel workbook, or PowerPoint presentation to the correct format handler with cross-cutting quality rules. Activates when archive, document, writer, report, present, visualize, or specify requests a binary document output. Summonable by any skill that needs to produce a stakeholder-ready deliverable.

Engineering capability gateway. Cross-cutting build standards, code quality floors, CI/CD requirements, and dependency hygiene that apply regardless of platform. Activates alongside any platform package. Catches quality gaps that platform engineering modules do not surface — test coverage floor, dependency freshness, documentation standards, code complexity limits, and universal CI requirements.

Experience capability gateway. Human-side validation for visual-facing projects — user research methods, usability testing, accessibility deep-dives (screen reader matrix, color blindness, cognitive, motor), and satisfaction measurement. Activates on visual targets where user research is declared in scope at P1. Requires explicit activation; Design gateway (accessibility floor) applies unconditionally.

Security capability gateway. Cross-cutting security concerns that apply regardless of platform. Activates on any target tagged with security-sensitive characteristics. Catches vulnerabilities that CLI, Web, and API/Service platform modules do not surface — git history exposure, cryptographic weaknesses, SAST findings, secrets in source, and cross-service trust failures.

Standalone grader. Evaluates primary output against Adversary counter-analysis and spec artifact. Issues ACCEPT / REVISE / ESCALATE verdict with score and revision guidance. Invokable directly by any caller — not only Reviewer. Do NOT invoke to generate the counter-case against an artifact — that is Adversary's job. Do NOT invoke when no Adversary output exists to grade against.

Pre-execution hallucination check. Verifies EARS/GWT requirements against actual repo state before Executor begins. Flags claims that cannot be confirmed in files. Runs after Decompose, before Executor wave 1. Produces ground receipt with verified/unverified/missing classifications.