v0.6.4

Fixed

• Copilot token usage now records (was always 0). Copilot writes its token totals only in
  the trailing session.shutdown events line, ~1s after agentStop — usage was sampled before
  it landed. read_usage now polls the transcript for a short grace, driven by a new per-profile
  usage_grace_s (8s for copilot, 0 elsewhere = read once).
• Copilot multi-turn reviews no longer stall. agentStop fires per response turn, so a
  parallel-subagent review ends several turns and tripped the global stop_without_result_nudges
  default of 1. New per-adapter floor (5 for copilot), overridable per stage via [adapter.review].

Added

• [adapter] usage_grace_s / stop_without_result_nudges (base + per-stage
  [adapter.dev|review|triage]), editable in the settings TUI. Unset = inherit the CLI profile's
  shipped default.

Changed

• Copilot docs. Pin a capable model — the free GPT-5 mini default silently skips steps in
  multi-step dev/review — and it's the Copilot CLI binary that's supported, not the VS Code
  extension.

v0.6.3

Fixed

• GitHub Copilot adapter (CLI 1.0.63). Turn-end is agentStop, not PascalCase Stop
  (which never fires) — every session previously read as a timeout. Remapped events, dropped
  the non-existent PreCompact, and the shared hook relay now reads camelCase payload keys
  (sessionId/transcriptPath). Probe mode sends its prompt verbatim so a skill-templating
  prompt_template no longer renders a missing-skill path that stalls the turn.

Added

• Copilot token accounting. New copilot-events usage_parser reads
  ~/.copilot/session-state/*/events.jsonl (data.modelMetrics.<model>.usage.*); the copilot
  profile is wired to it (was usage_parser = "none").

v0.6.2

Added

• bmad-auto probe-adapter (alias collect-adapter-data). A self-service command that
  collects and sanitizes everything needed to finalize a CLI adapter profile — the hook payload
  shape, transcript location/format, and token-usage schema for a usage_parser — so a user of
  any coding CLI can paste back a clean, content-free report. A default zero-launch scan reads
  on-disk conventions; opt-in --probe does a live capture in an ephemeral workspace. All output
  passes through one audited PII sanitizer (token counts and field names survive; paths, prose, and
  emails are redacted).
• GitHub Copilot CLI profile. Bundled copilot profile (Copilot CLI ≥ 2026-02): -i
  interactive launch, VS Code-compatible Stop hook, --allow-all-tools for unattended runs.
  Still pending live E2E and a usage_parser — probe-adapter captures the token schema to write
  one.

Docs

• Adapter authoring guide. New adapter authoring guide
  walks through finalizing a CLI profile with probe-adapter (scan vs probe, the PII model, and
  the parser-writing loop); probe-adapter is added to both command references.

v0.6.1

Added

• Short run refs (Docker-style). Every command that takes a run id (status, attach,
  resume, resolve, stop, delete, archive) now accepts a partial — the tail after the
  last - (e.g. a1b2, or as few chars as stay unique). Full ids still work; an ambiguous ref
  fails listing the candidates. New bmad-auto list (alias ls) prints each run/sweep with its
  short ref, type, and status.
• Flexible --story selection. bmad-auto run --story now takes more than the exact full
  key: an epic+number (--epic 3 --story 1, --story 3-1, or --story 3.1) or a slug fragment
  (--story user-auth). Full keys still work. Mismatches are caught before the run launches with a
  targeted error — no match, ambiguous slug, or matched-but-not-actionable.

v0.6.0

Fixed

• Rollback no longer wipes non-automator files. A failed in-place attempt previously ran
  git reset --hard + a blanket git clean -fd over the whole checkout, which could delete a
  project's _bmad-output/ and any other untracked files (only .automator/ and two artifact
  subdirs were spared). The orchestrator now never runs a blanket git clean: it reverts the
  attempt's tracked changes and removes only the untracked files that run created, preserving
  pre-existing untracked files and the entire _bmad-output/ tree.

Changed

• Auto-rollback is now opt-in ([scm] rollback_on_failure, default off). With it off the
  orchestrator never touches your working tree on a failed attempt — it pauses the run with bold
  manual-recovery instructions (back up untracked files → git reset --hard <baseline> → restore).
  Turn it on for the safe automatic rollback above (it discards the attempt's uncommitted work, so
  it warns when it fires). Worktree isolation (scm.isolation = "worktree") sidesteps this entirely.