Skip to content

CHANGELOG.md

Latest commit

 

History

History
94 lines (80 loc) · 4.98 KB

CHANGELOG.md

File metadata and controls

94 lines (80 loc) · 4.98 KB
Raw

Changelog

All notable changes to Copilot-swarm (CSW) are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[0.1.3] - 2026-07-12

Fixed

  • Linked the bundled csw-plan workflow reference from SKILL.md so Awesome Copilot's orphan-file quality gate can verify that every shipped skill asset is reachable.

[0.1.2] - 2026-07-12

Changed

  • Migrated active delegation guidance to Copilot CLI's native task subagents, /fleet parallel execution, and /tasks oversight/cancellation; removed the custom dispatcher from the shipped plugin.
  • Replaced generated install permission profiles with simple install/dry-run flows and host-native permission guidance. Investigation safety now requires host tool restrictions, while writing workers require isolated worktrees.
  • Hardened completion evidence around machine-generated verify and artifact receipts. Free-text evidence can report failure or pending context but cannot pass a criterion.
  • Limited continuation to the root agentStop surface. Missing, malformed, empty, stale, completed, and safe-mode state intentionally fail open rather than trapping the host.
  • Documented the accepted receipt boundary: Git freshness excludes ignored inputs, non-git verification has no freshness guarantee, and timeout/cancel process-tree cleanup is best-effort. Verified commands must be approved and non-daemonizing; ignored inputs and cleanup require explicit receipts.

[0.1.1] - 2026-06-19

Added

  • Release preflight checks for version lockstep, README-linked package payloads, and exclusion of local CSW evidence artifacts from npm packages.
  • Deterministic redaction for CSW runtime state, ledger, hook snippets, and MCP worker summaries before they are persisted or surfaced.
  • Safe mode (CSW_SAFE_MODE=1) and stale-state fail-open behavior for continuation, disabling hook context emission/auditing, and dispatch escape hatches.
  • Install-time permission profiles (safe, balanced, full, none) for npx and local csw install flows, including dry-run reporting and MCP tool/profile wiring.

Changed

  • npm prepublishOnly now runs tests, forbidden-token scans, release checks, and a dry-run package build before any human-gated publish attempt.
  • The npm payload now includes README-linked local assets/docs (cover.png, README-Ko-KR.md, and docs/).

[0.1.0] - 2026-06-07

First release: a native GitHub Copilot CLI plugin for parallel delegation and an evidence-gated plan → execute → review workflow.

Added

  • Native plugin (.plugin/plugin.json) registering skills, custom agents, hooks, and an MCP server. Installs via copilot plugin install (or csw install).
  • Swarm dispatch MCP (mcp/dispatch/) — a zero-dependency MCP stdio server that restores model-callable parallel delegation by orchestrating parallel copilot -p workers: tools dispatch, code_search, research. Bounded concurrency, per-worker timeout, failure isolation, and a recursion-depth guard.
  • Worker roster — six focused custom agents: explorer, researcher, planner, gap-analyst, plan-reviewer, verifier (namespaced copilot-swarm:*).
  • Durable evidence-gated goal runtime (runtime/, bin/csw-runtime.mjs) — machine success criteria (C0NN | channel: | test: | scenario:), atomic .csw/ state + append-only JSONL ledger, a completion oracle (every criterion pass with evidence and zero open blockers), a steering guard that refuses gate-weakening, and a clear escape hatch.
  • Skills — swarm (parallel delegation), csw-plan (explore-first planning with an approval gate), csw-work (disciplined execution), csw-review (multi-lane all-or-nothing review), and csw-loop — typing the csw keyword runs the full evidence-bound autonomous loop (goal → plan → test-first execute → QA → review → oracle), declared in the session-start doctrine so it activates on the keyword.
  • Hooks — sessionStart doctrine injection (the always-on instruction vehicle), userPromptSubmitted steering audit, postToolUse comment-checker (anti AI-slop), and agentStop/subagentStop continuation gate (force-continue until the oracle passes).
  • Install UX — bin/csw CLI with an ANSI banner, theme presets, and status/install/doctor commands.
  • HUD status line — bin/csw-statusline.mjs for Copilot CLI's statusLine surface: shows the active goal's criteria progress, open blockers, and objective below the input (nothing when no goal is active). Enable via csw hud.
  • Release-cleanliness oracle — scripts/scan-forbidden.mjs gating the tracked, packable, and npm pack tarball surfaces.

Notes

  • No telemetry, no call-home, no external runtime dependencies.
  • LSP is delegated to Copilot's native support; see docs/supporting-components.md for the port/keep/skip rationale.