mbourmaud/sillage-workflow

v0.2.0MIT

A human-governed, resumable engineering workflow for coding agents.

Changelog

All notable changes to Sillage are documented here. This project follows Keep a Changelog and uses Semantic Versioning.

Full workflow kernel with decision-bound review and handoff artifacts.
Read-only context and status views for fresh-agent cold starts.
Explicit, atomic transition --write for local task records, with optimistic concurrency and symlink protections.
Initial orchestration skill, compact workflow templates, and a complete executable pilot.
A responsive public website with workflow, installation, and release pages.
Full-workflow installation and deliberate-update guidance for Agent Skills clients.
Provider-neutral execution profiles for matching reasoning capability and effort to each workflow stage without naming a model.
Provider-neutral delegation requests for bounded parent or subagent work, with explicit isolation and return packets.
Codex adapter profiles for decision research, isolated implementation, and read-only review.
A canonical provenance policy documenting inspirations, naming boundaries, and MIT attribution rules.
A skills-first V1 suite with explicit using-sillage entry and compact Markdown task-card guidance.
Language-neutral architecture, testing, interface, systems, security, and platform lenses with a shared engineering doctrine for pressure, seams, patterns, failure, compatibility, and proof.

Release validation now requires a versioned changelog section before a tag can publish.
Specialist lenses are progressive and non-prescriptive: DDD, SOLID, Clean Architecture, patterns, HTTP, networking, security, and platform practices are selected by observed risk rather than imposed on every project.

Human-governed, forge-neutral task lifecycle and JSON contracts.
Canonical project entry points and deterministic doctor, digest, and read-only transition commands.
Agent Plugins 1.0 manifests for Codex and Claude Code distribution.
Initial evidence-led research Agent Skill with evaluation prompts and a real installation pilot.