Releases · krzemienski/shannon

Shannon v1.2.0

Second stable release. Builds on v1.0.0 with deep planning, parallel wave execution, and the evidence-gated forge pipeline.

DeepPlan + wave-execution — multi-phase plans decomposed into dependency-ordered parallel waves with between-wave synthesis gates
/shannon:plan-deep — true alias of plan --mode deep (phantom flags removed; only real plan.md flags exposed)
Forge pipeline — codebase-analysis → docs-research → planning → oracle plan-review → execute → validation → evidence-indexing → 3-reviewer consensus → 3-oracle quorum → completion-gate
skills10x harness — self-contained v1-native skill activation harness; registration derived transitively from authoritative Skill tool_use
Dynamic Workflows awareness + real Agent SDK load probe (scripts/harness/load_check.py)
doctor.py reads version live from the manifest

36 skills · 11 agents · 22 commands · 9 hooks

No mocks, no stubs, no test files — real-system evidence or a refusal. Enforcement hooks block .test.* writes at write-time.

First stable release of Shannon — a Claude Code plugin for high-stakes multi-stage agentic work.

Surface: 33 skills · 9 agents · 20 commands · 7 hooks.

Five pillars:

Embedded sub-agent skills — sub-agents carry their skills inline; spawning never silently fails to load.
Orchestration — single-message multi-Task dispatch (sequential / parallel / competitive), with escalation to Dynamic Workflows for unbounded fan-out.
Iron Rule validation — real-system evidence on disk. No mocks, no stubs, no test files.
Meta-judge consensus — rubric YAML generated before any judge runs; hidden thresholds; debate on disagreement.
Self-instrumented — scripts/doctor.py (10/10 contract checks) and scripts/harness/load_check.py (real Agent SDK load probe, inherits CLI auth).

Verified at release: doctor 10/10 PASS; SDK load probe PASS (all 20 commands addressable when the plugin loads).

Install:

/plugin marketplace add krzemienski/shannon
/plugin install shannon@shannon