Fix Wizard pipeline hooks and support additive phases

[nicolehaugen]

Summary

• Normalize extension hook targets and render a single connector between pipeline steps, with hook tooltips listing the dispatched commands.
• Add an agent-facing addPipelinePhases canvas action that inserts installed, manually runnable commands without removing or reordering existing phases; persist and broadcast changes.
• Ignore generated specs/ artifacts so sample Wizard runs do not enter repository changes.

Validation

• Wizard suite passed 287/287 tests after the pipeline changes.
• Rebasing onto the latest origin/main completed cleanly; git diff --check passed.

Generated specifications and Wizard session state are not included.

[Copilot]

Copilot review overview

🟡 Changes recommended

Mixed-anchor additions can be ordered incorrectly, and concurrent state updates may be overwritten.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1 · 1

Open (2)

• Make state read-modify-write atomic · New
• Preserve ordering across transitive insertion anchors · New

What changed in this PR

Adds reliable hook visualization and agent-driven additive pipeline phases while preserving Wizard state.

Changes:

• Groups hook commands by lifecycle phase with simplified connectors.
• Adds, persists, and tests addPipelinePhases.
• Normalizes hook targets and ignores generated specifications.

File	Description
.gitignore	Ignores generated specifications.
README.md	Documents the new action.
extension.mjs	Registers pipeline actions.
composition-apply.mjs	Normalizes hook targets.
actions/​pipeline.mjs	Implements additive phases.
phase-card.js	Groups and renders hook steps.
stepper.css	Removes obsolete hook separators.
composition.test.mjs	Tests hook normalization/rendering.
pipeline-action.test.mjs	Tests pipeline additions and persistence.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🟢 Approval recommended

The additive pipeline and hook behavior is coherently implemented and covered by focused tests.

Review effort: Balanced
Findings: 1

Open (1)

• Preserve ordering across transitive insertion anchors

Resolved since last review (1)

• Make state read-modify-write atomic

[Copilot]

Copilot review overview

🟡 Changes recommended

Partial retries can place new phases before an already-present transitive anchor chain.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1

Open (1)

• Partial retries misorder phases with transitive anchor chains · New

Resolved since last review (1)

• Preserve ordering across transitive insertion anchors

[Copilot]

Copilot review overview

🔵 Needs a closer look

Hook normalization retains a stale top-level target that can override the corrected binding target.

Review effort: Balanced
Findings: None

Resolved since last review (1)

• Partial retries misorder phases with transitive anchor chains

Previously missed (1)

In code that hasn't changed since last review

Normalize artifact targetCommand to the selected binding target

plugins/​spec-kit-copilot-wizard/​extensions/​speckit-wizard-canvas/​canvas-runtime/​composition-apply.mjs:152

This keeps a stale top-level artifact.targetCommand when the binding-derived target wins. resolveHooksForCommand prefers a.targetCommand over each binding's targetCommand (pipeline/active-artifacts.mjs:229), so the conflicting legacy case exercised below still resolves and displays the old command despite the normalized ID/bindings. Normalize the top-level field to the selected target as well.

[nicolehaugen]

Won't fix (hook target normalization edge case). The reported mismatch requires a hook artifact whose top-level targetCommand conflicts with its binding's targetCommand. The current deterministic composition assembler creates hook artifacts with matching ID/binding targets and does not set a competing top-level target, so ordinary extension installation and refresh do not produce this shape. We have not identified a normal Wizard workflow that generates the conflict; it is covered only by a synthetic inconsistent/legacy input in the test.

If such a record is supplied from stale or externally constructed metadata, the Wizard could display the old hook command because resolveHooksForCommand prefers the top-level field. This is a display/attribution issue, not evidence that the actual hook executor runs a different command or that artifacts are deleted. Given the narrow trigger, we are deferring the defensive normalization change for this PR. Revisit if a real producer or user report demonstrates the conflicting shape.

[Copilot]

Copilot review overview

🟢 Approval recommended

The implementation is consistent, release metadata is synchronized, and the changed behavior has focused regression coverage.

Review effort: Balanced
Findings: None