v0.10.0

Added

• /craftsman:discover <vague idea>: a requirements-discovery interview that turns an
  underspecified request into a numbered, testable spec. It exists because the shallow end of
  ideation-first was the only option — 3–5 questions and a three-line brief, with no way to invoke
  it deliberately, so a substantially open request ("build a notifications system") got the same
  treatment as a merely fuzzy one.

  The work is split across two surfaces because of a hard constraint: a subagent cannot conduct an
  interview. It receives one cold prompt and returns one result, which is why orchestrator already
  has to bounce underspecified features back to its caller. So the interview runs in the main thread
  as a command, and only the half a subagent is genuinely good at — reading the codebase — is
  delegated.

• requirements-analyst agent (read-only: Read/Glob/Grep, Opus): drafts the clarifying questions a
  request leaves unanswered, grounded in prior art and existing abstractions rather than a generic
  checklist. Every question carries a file:line or an explicit "gap" label, plus an assumption
  default so the interview never stalls on a declined question; questions the code already settles
  come back as constraints instead. Capped at 10–20 questions — past that, it is asking things the
  codebase already answers. It never asks the user anything itself and never writes files.

Changed

• ideation-first: now runs at two depths. Quick is unchanged and still the default (3–5
  questions, one per turn, Scope brief). Deep grounds 10–20 questions in the analyst's dossier and
  asks them in 3–4 sequenced rounds of 2–4 questions via AskUserQuestion — a deliberate
  divergence from the one-question-per-turn rule, which is right for five questions and punishing for
  fifteen. What the rule actually protects is that later questions adapt to earlier answers, and
  sequenced rounds preserve that; batching all fifteen at once does not.

  Deep mode ends by writing a numbered requirements spec (R1, R2, … each with an acceptance
  criterion, so tester and reviewer can trace coverage back to a requirement) at
  docs/requirements/<slug>.md — proposed, shown in full, and written only on confirmation, per the
  same ask-before-writing rule /craftsman:init follows. The Scope brief is still emitted inline
  at both depths and remains the machine handoff the planner and the orchestrator's ideation gate
  consume; the spec file is additive, and the brief gained a Spec: line pointing at it.

  New explicit rule: expand the questions, bound the scope. Deep mode raises error states, empty
  states, migration, and offline behavior — then still recommends the smallest build that satisfies
  the request, with everything surfaced but unagreed recorded under Out of scope. A deep interview
  that ends with a bigger scope than a quick one has failed; it should end with the same scope, held
  with more confidence and a longer out-of-scope list. Without this, deep mode would have quietly
  contradicted smallest-change-first.

• orchestrator: the ideation gate now accepts either an inline Scope brief or a path to a
  requirements spec, and instructs the planner to read the spec when one is named. Availability table
  gains requirements-analyst.

• /craftsman:orchestrate: the ideation-gate step routes substantially open requests to deep mode
  rather than a three-line brief.

• userpromptsubmit-task-router: the feature-shaped nudge names /craftsman:discover for
  substantially open requests. Still one nudge per prompt; no new case branch.

Fixed

• session-start announced "Seven skills" and "Five commands" — both already wrong before this
  release (8 and 6), and both invisible to CI: the count validator skips any line containing
  "commands" when checking skills, and only matches the literal phrase "N slash commands" for
  commands. The whole reminder is a single line, so both slipped through every check. The skill list
  was also missing ui-craft and the command list was missing /craftsman:orchestrate — a session
  therefore started without being told two components existed.

v0.9.0

Added

• /craftsman:orchestrate <task>: a slash command for the full pipeline. The orchestrator was the
  only major entry point without one, so it never appeared in the / menu — every other command
  (init, quick, doctor, review, security) did. It runs the orchestrator's discipline in
  the main thread rather than dispatching the agent as a subagent, for two reasons:
  orchestration is subagent dispatch and nesting that is unreliable, and the ideation gate needs
  interactive Q&A that a subagent cannot hold (as a subagent, orchestrator must bounce
  underspecified features back to its caller). @craftsman:orchestrator still works for running
  the pipeline in its own context.

Fixed

• Agent invocation is documented as @craftsman:<name> throughout. Plugin agents are namespaced,
  so the bare @<name> form the docs taught — @orchestrator in both READMEs, docs/use-cases.md,
  and the ideation-first skill — matches nothing in the agent picker, making the orchestrator look
  like it wasn't installed. Note: craftsman-demo.svg still shows @orchestrator in its rendered
  terminal text; correcting it requires regenerating the asset.

v0.8.0

Added

• Automatic skill-routing hooks so craftsman skills surface from task shape instead of a manual
  check. Four new event-driven hooks, all self-contained (they reference only craftsman:* skills
  and this plugin's own agents, never another plugin) and quiet by default (a nudge fires only on a
  real signal): userpromptsubmit-task-router (routes the prompt to the matching agent/skill by
  task shape — bug, feature, refactor, security, release, CI, testing, docs, version-control),
  pretooluse-git-guard (commit-craft hint on git commit/push/tag & gh pr),
  pretooluse-edit-guard (file-pattern skill hints for KNOWN_ISSUES.md/test/workflow files, and
  drops the session marker the close-out hook reads), and stop-closeout (nudges
  caveats-and-status once after real edits, then self-deletes its marker — loop-safe via both the
  marker and stop_hook_active).
• New hook events wired in hooks.json: UserPromptSubmit and Stop (previously only
  SessionStart + PreToolUse).

Changed

• orchestrator: gate and repair hardening. Adds a handoff contract — every dispatched agent
  must close with STATUS/FILES CHANGED/VERIFIED BY/EVIDENCE/CAVEATS, and gates are checked
  against that block rather than the agent's prose, so a missing evidence line now fails the gate
  instead of passing on confidence. The repair loop escalates rather than retrying: attempt 2
  routes to a different agent (debugger for build/test failures, planner for review findings)
  before re-implementing. The parallelism rule now forbids two file-writing agents running
  concurrently (the old "different output artifacts" wording permitted it) and states that parallel
  dispatch means multiple Agent calls in one message. Also: one-sentence scope restatement and a
  new-dependency stop before pipeline selection, a bugfix requirement that the regression test fail
  before the fix and pass after, Files changed / Verification run lines on the report, and an
  explicit "never commit, push, branch, or open a PR unless asked" constraint.
• session-start: added a proactive lead-in to check environment-quirks.md before retrying a
  failed command, and to prefer graphify for navigation when graphify-out/graph.json exists.
• run-hook.cmd: reworded to drop the superpowers-plugin attribution — the dispatcher and the
  whole plugin depend on no other plugin being installed.
• README: new "Automatic skill routing (hooks)" section documenting the six hooks, plus the four
  routing ideas (skill chaining, parallel analysis, negative filtering, contextual caching) that
  are intentionally model/orchestrator behaviors rather than hooks.

v0.7.0

Added

• /craftsman:doctor --fix: opt-in remediation for the three detected silent-failure conditions.
  Default (no-arg) behavior is unchanged — diagnostic only. With --fix, each repair is shown and
  confirmed (y/n) before it runs, then the check is re-run to verify: Git Bash via
  winget install Git.Git (lands on the standard path the hook dispatcher already probes, so no
  PATH edit), graphify graph via graphify ., and the Windows python3 Store-stub via
  winget install Python.Python.3.12 (flagged as a partial fix — the Store app-execution alias may
  still need disabling). Falls back to manual install links where winget is unavailable.

v0.5.1

Added

• Terminal-style demo image (craftsman-demo.svg) on the root README showing an orchestrator
  bugfix run end to end.
• CONTRIBUTING.md: local testing via --plugin-dir, validation requirements, component
  conventions, run-hook.cmd polyglot editing rules, and versioning/changelog process.
• docs/use-cases.md: four worked scenarios (minimal diffs, recurring-bug recall via graphify,
  ask-first scaffolding, honest completion).
• GitHub issue templates (bug report, feature request) and a Discussions contact link.

Changed

• Restructured both READMEs to lead with the value proposition: a "Why craftsman" problem/solution
  table, a 30-second start, and status badges (CI, version, license). Reference content unchanged;
  the plugin README's Quick start section merged into the new 30-second start.

Full changelog: https://github.com/bufferBrew/craftsman/blob/main/CHANGELOG.md