krzemienski-shannon-framework

Plugin krzemienski-shannon-framework from marketplace cascade-ai-ml-agents-agent-framework in joshuarweaver/cascade-ai-ml-agents-agent-framework. Commands read from the repo's manifest — 1 month ago.

Skills (31)

• shannon-execution-verifier

  Comprehensive post-build verification of Shannon Framework's application outputs using three-layer methodology: Flow Verification (execution trace analysis), Artifact Verification (physical output inspection), and Functional Verification (runtime testing). Verifies Shannon built production-ready app

• testing-claude-plugins-with-python-sdk

  Use when testing Claude Code plugins programmatically, building SDK applications, encountering plugin loading failures (plugins array empty), getting AttributeError on message.type, or needing verified Claude Agents SDK Python patterns - provides complete reference for correct message handling (isin

• codebase-analysis

  Build repo-wide context before any modification. File inventory, module map, dependency manifests, hot-path identification. Read-only.

• completion-gate

  Evaluate the completion gate. Refuses on any missing MSC. Reads evidence tree; emits machine-readable report.json. No override flag.

• documentation-research

  Fetch and cite current upstream documentation for every external dependency in scope. ISO-8601-stamped sources. Refuse memory-only references.

• evidence-indexing

  Maintain README.md (purpose) and INDEX.md (artifact enumeration) in every evidence directory. Refuse to leave any evidence dir un-indexed at gate completion.

• oracle-review

  Pre-execution plan-review AND post-execution evidence-audit. Quorum-gated. Spawns ≥3 isolated oracle agents; ≥2 APPROVE required.

• refusal-discipline

  When evidence is missing, write a structured REFUSAL.md and stop. No override, no force-complete. Refusal is a feature.

• dispatch-parallel

  Launch N parallel sub-agents with model selection, quality-focused prompting, meta-judge verification. Used in parallel, sequential, and competitive modes.

• judge

  LLM-as-judge evaluation of subagent outputs. Debate mode resolves ties via cross-judge argument.

• tree-of-thoughts

  Systematic exploration + pruning + expansion via Tree of Thoughts methodology. Branch on candidate sub-thoughts; meta-judge prunes.

• observability-report

  Shannon-specific log reader. Powers /shannon:trace, /shannon:doctor, retrospective session analysis. Reads hooks.jsonl, hook-errors.jsonl, session JSONLs.

• autopilot-runner

  Refusal-driven retry orchestration for /shannon:autopilot. Wraps /shannon:cook with REFUSAL.md parsing and remediation prompt synthesis.

• loop-runner

  Iterative do-verify-reflect loop until verify skill PASSes or max-iter exhausted. Backs /shannon:loop.

• team-coordinator

  Multi-teammate orchestration. Maintains shared TaskList, assigns work by file-ownership glob, synthesizes findings, blocks Stop on incomplete tasks.

• create-validation-plan

  Multi-phase project plan generator that injects functional validation gates with evidence requirements. Output: phase plans with gates per real-system PASS criterion.

• plan-author

  Linear hierarchical plan author. plan.md + phase-NN.md files with measurable success criteria and validation gates per phase.

• plan-converge

  Convergence orchestrator. Iterative plan-author → red-teamer → plan-author until red-teamer reports ≤1 BLOCKING.

• plan-tournament

  Tournament orchestrator. Spawns N plan-author candidates with distinct perspectives, red-teamer per candidate, tournament-judge selects winner.

• research-validation

  Gather standards, best practices, applicable criteria before validation planning. Maps standards (WCAG, HIG, security) to Shannon validation skills.

• sequential-analysis

  Structured multi-step reasoning with revision capability. For complex problems, root-cause investigation, ambiguous scope.

• critique

  Adversarial critique of current work. Find holes, name what's missing, identify where assumptions break.

• memorize

  Extract a learned pattern from session and save to memory. Cross-link with [[name]] to related memories.

• plan-do-check-act

  Iterative PDCA cycle for systematic experimentation and continuous improvement.

• reflect

  Self-refinement pass on prior output. Read last turn, identify gaps, propose specific fixes.

• root-cause-tracing

  Five-whys + cause-and-effect diagram + analysis report. Drives /shannon:why.

• consensus-engine

  Multi-validator agreement gate. N independent validators against same feature; synthesize confidence-scored verdict.

• evidence-gate

  5-question checklist applied before any completion claim. Any 'no' refuses completion.

• functional-validation

  End-to-end no-mocks validation through real system execution. Build, run, exercise UI, capture evidence, apply gate validation.

• no-fakes-discipline

  Refuses circumvention of validation via fake substitutes. Reinforces the iron rule that the real system must be exercised.

• visual-inspection

  Systematic visual QA for UI screenshots. iOS HIG, web WCAG 2.2, cross-platform heuristics.

Slash commands (27)

• /krzemienski-shannon-framework:audit-completion

  Read evidence tree for a forge run, print MSC table, reviewer consensus, oracle quorum, overall verdict. Read-only.

• /krzemienski-shannon-framework:audit

  Read-only audit with severity classification. Screen, app, session, or drift scopes. No edits.

• /krzemienski-shannon-framework:autopilot

  Refusal-driven retry loop around /shannon:cook. Preserves Crucible's REFUSAL.md pattern.

• /krzemienski-shannon-framework:cook

  Execute a plan or task brief end-to-end with iron-rule validation and evidence gates. Wraps planning, execution, functional validation, and completion gate.

• /krzemienski-shannon-framework:disable

  Opt this project OUT of Shannon enforcement. Writes .shannon/disabled and removes .shannon/active. Sticky — survives until /shannon:enable --force.

• /krzemienski-shannon-framework:dispatch-competitive

  N parallel subagents on same task; multi-judge evaluation; evidence-based synthesis.

• /krzemienski-shannon-framework:dispatch-parallel

  Parallel subagents on independent tasks. Meta-judge + LLM-as-judge verification per sadd's pattern.

• /krzemienski-shannon-framework:dispatch

  Sequential subagent chain. Each subagent gets IRON-RULE inject. Judge confirms each output before next dispatches.

• /krzemienski-shannon-framework:doctor

  Health check. Validates plugin manifest, hooks.json registration, settings.json sanity, log dirs writable, marketplace declarations.

• /krzemienski-shannon-framework:enable

  Opt this project into Shannon enforcement by writing the .shannon/active sentinel. Hooks become no-ops in projects that have not opted in.

• /krzemienski-shannon-framework:fix

  Scout → debug → implement → revalidate. Three-strike cap per VF forge convention. Each attempt writes fresh evidence directory.

• /krzemienski-shannon-framework:forge

  Evidence-gated end-to-end completion pipeline. Codebase-analysis → docs-research → planning → oracle plan review → execute → validation → evidence-indexing → reviewer consensus → oracle quorum → completion-gate. Refuses on cited blocker.

• /krzemienski-shannon-framework:install

  Idempotent setup. Writes marketplace declaration; runs uninstall-others.sh (after y/N prompt) unless --parallel. Verifies via doctor at end. Local marketplace only (UQ-CMD-3).

• /krzemienski-shannon-framework:loop

  Self-referential do-verify-reflect loop until goal verified or max iterations hit. Absorbs all ralph variants.

• /krzemienski-shannon-framework:plan-converge

  Iterative plan refine → critique → revise. Converges when red-teamer returns ≤1 BLOCKING.

• /krzemienski-shannon-framework:plan-deep

  Deepest-plan style. Wraps plan-tournament + plan-converge with gate injection. Consensus + gates + synthesis.

• /krzemienski-shannon-framework:plan-tournament

  Tournament-style plan generation. N parallel plan-author candidates with distinct perspectives; red-teamer per candidate; tournament-judge selects winner.

• /krzemienski-shannon-framework:plan

  Linear hierarchical plan author. Drops plans/{date-prefix}-{slug}/{plan.md + phase-NN.md files}. Validation phase included by default.

• /krzemienski-shannon-framework:prd

  Structured Product Requirement Document authoring. Interview-driven sections (Why / What / How / Success Metrics / Risks).

• /krzemienski-shannon-framework:reflect

  Self-refinement pass. Modes: self (read last turn, propose fix), critique (spawn critic agent), memorize (save learned pattern).

• /krzemienski-shannon-framework:research

  Parallel researcher agents on sub-topics; aggregator writes cited research summary. Standalone command, not subcommand (UQ-CMD-2).

• /krzemienski-shannon-framework:resume

  Inspect evidence tree, identify last completed phase, resume from the next missing phase artifact. Evidence-tree-as-state.

• /krzemienski-shannon-framework:retro

  Mine session JSONLs over N days; aggregate decisions, lessons, gotchas; write retrospective report.

• /krzemienski-shannon-framework:team

  Spawn coordinated multi-agent teammates with shared TaskList, charter-driven roles, and stop-task semantics. Replaces OMC team and VF validate-team orchestrators.

• /krzemienski-shannon-framework:trace

  Read session JSONL. Emit chronological timeline of hooks fired, skills invoked, tools called. Proves which Shannon hooks fire in real sessions.

• /krzemienski-shannon-framework:validate

  Detect platform, run validation journeys against the real system, capture evidence, emit cited PASS/FAIL verdicts. Replaces VF validate family and deepest-plan validate.

• /krzemienski-shannon-framework:why

  Five-whys + root-cause-tracing + cause-and-effect diagram. Output root-cause analysis report.

Subagents (10)

• coordinator

  Multi-teammate coordinator. Maintains shared TaskList, assigns tasks by file-ownership glob, synthesizes findings, resolves conflicts. Lead role for /shannon:team.

• critic

  Adversarial critique of work artifact. Surfaces what's missing, what's wrong, where assumptions break.

• dispatch-judge

  Meta-judge for /shannon:dispatch-* commands. Evaluates parallel subagent outputs, scores them, emits winner with cited reasoning.

• executor

  Execute a phase plan against the codebase. Reads phase-NN.md, makes edits, runs validation per phase gates. NEVER spawns nested teams. NEVER creates test files.

• oracle

  Final-gate Oracle. Reviews a plan (pre-execution) OR a completed evidence package (post-execution) in isolated context. Distinct from reviewer per UQ-SA-2. Emits APPROVE / REFUSE with cited blocker.

• planner

  Authors implementation plans. Reads codebase, identifies phases, writes measurable success criteria. NEVER edits source code.

• red-teamer

  Dispatch hostile review with --lens arg (security|scope|evidence|failure-modes). Single agent, multiple lenses (per UQ-SA-1).

• researcher

  Researches a sub-topic; outputs cited summary. Used by /shannon:research and /shannon:plan-deep.

• reviewer

  Post-implementation review for correctness, security, performance, style. Cites evidence per finding. Language-agnostic; per-language bias comes from project CLAUDE.md, not separate agents.

• validator

  Run validation suite per platform. Detect platform, execute journeys, capture evidence, emit per-journey verdict. Routes to platform-specific dispatch.

Hooks (16)

• SessionStart

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/session-context-inject.cjs"

• UserPromptSubmit

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/skill-activation-check.js"

• PreToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/block-fab-files.js"

• PreToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/read-before-edit.js"

• PreToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/plan-before-execute.js"

• PreToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/evidence-gate-reminder.js"

• PreToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/subagent-governance-inject.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/validation-not-compilation.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/validation-skill-tripwire.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/completion-claim-validator.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/fab-pattern-detection.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/evidence-quality-check.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/task-list-tracker.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/hooks-fired-log.js"

• PostToolUse

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/context-threshold-warn.js"

• Stop

  PATH="/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:$PATH" node "${CLAUDE_PLUGIN_ROOT}/hooks/stop-task-semantics.js"

Similar plugins