Conversation
…ck-list Update parseFrontmatterSkills() in skill-references.test.ts to parse YAML block-list format (`- devflow:name` entries) instead of comma- separated strings. Convert all 12 shared agents to use the new format. Block-list YAML is the correct frontmatter format for multi-value arrays and enables Claude Code to preload skills into subagent context at spawn time via the platform's skill injection mechanism. Co-Authored-By: Claude <noreply@anthropic.com>
…er mode skills - designer.md: preload gap-analysis and design-review into frontmatter block-list (eliminating the conditional Read of mode skills at runtime); update Modes table header from "Read this first" to "preloaded" - coder.md: replace install-path Read instructions with Skill tool invocations for domain skill loading; fail BLOCKED on Skill error - reviewer.md: replace install-path Read instruction with Skill tool invocation for focus skill loading; update Focus Areas table to use devflow:X references instead of ~/.claude/skills paths - tests: update parseReviewerFocusAreas() regex and coder domain skill check to match new formats; lower Format 3 install-path threshold to >=0 since agents now use Skill tool invocations Co-Authored-By: Claude <noreply@anthropic.com>
Add getLatestSubagentPreloadedSkills() to integration test helpers —
walks ~/.claude/projects/{encoded-path}/*/subagents/agent-*.jsonl to
find the most recent subagent transcript after a given timestamp and
extracts skill names from <command-name> tags in the first user message.
Add subagent-skill-preload.test.ts with 6 smoke tests verifying that
skills declared in agent frontmatter block-lists are injected into
subagent context at spawn time. Tests are guarded by isClaudeAvailable()
and run under npm run test:integration only (not npm test).
Assertions validated:
- Simplifier: software-design + worktree-support, no apply-knowledge
- Scrutinizer: quality-gates, software-design, worktree-support, apply-knowledge
- Reviewer: review-methodology, worktree-support, apply-knowledge
- Coder: all 8 core skills
- Designer: worktree-support, apply-knowledge, gap-analysis, design-review
- Git: git + worktree-support
Co-Authored-By: Claude <noreply@anthropic.com>
- Replace __dirname with import.meta.dirname for ESM consistency - Remove debug console.log statements from integration tests - Remove dead code (empty if-block for tool_result events) - Reuse CLASSIFICATION_PATTERN constant in hasDevFlowBranding - Remove always-true assertion (count >= 0) - Use explicit length check instead of toBeTruthy for string assertion
Domain skills in Coder and focus skills in Reviewer are optional enhancements — missing optional plugins should not block task completion. - Coder step 2: skip failed domain skill invocations and continue rather than reporting BLOCKED; remove misleading "First action —" prefix - Reviewer step 1: fall back to built-in knowledge if focus skill load fails rather than reporting BLOCKED - Both: add inline rationale comment explaining why dynamic loading is used (unbounded domain/focus set vs preloading all variants) - Restore hasDevFlowBranding export in integration helpers (accidentally deleted in prior working-tree edit, breaking ambient unit tests) Co-Authored-By: Claude <noreply@anthropic.com>
- Fix timing race: use >= instead of > for mtime comparison so transcripts written at the same millisecond as the marker are not dropped (Issue 1) - Decompose getLatestSubagentPreloadedSkills into three focused functions: findRecentSubagentTranscripts, parsePreloadedSkills, and a thin orchestrator (Issue 2) - Add COUPLING comment documenting dependency on Claude Code's internal transcript format (Issue 3) - Bound session directory scan to most recent 20 entries to avoid O(N) walk on machines with hundreds of sessions (Issue 4) - Add unknown type annotation and type guards for both JSON.parse call sites to satisfy TypeScript strict null checking (Issue 5) - Document hasDevFlowBranding as intentional alias of hasClassification via @see JSDoc — function is used in ambient.test.ts (Issue 6) Co-Authored-By: Claude <noreply@anthropic.com>
Add non-empty guard before arrayContaining in all 6 subagent skill preload integration tests so a missing transcript produces a clear diagnostic instead of a cryptic arrayContaining failure. Add a new Format 2 test that asserts every shared agent declares at least one skill in frontmatter, guarding against silent regressions where parseFrontmatterSkills returns [] for a malformed block. Co-Authored-By: Claude <noreply@anthropic.com>
…ious async - Collapse 6 repetitive test scaffolds into shared helper with centralized guard assertion - Remove async keyword from 5 synchronous test functions (sync I/O by design)
- Rename Promise callback parameter from `resolve` to `done` in helpers.ts to avoid shadowing the path.resolve import - Reorder simplifier.md and skimmer.md frontmatter so model: precedes skills:, matching the other 10 agents
- Add runClaudeAndWait helper for subagent tests (waits for completion instead of early-exit on Skill detection) - Use --dangerously-skip-permissions so subagent tools execute in non-interactive -p mode - Fix parsePreloadedSkills to scan all isMeta user messages (skills are injected as separate messages, not in the first prompt message) - Add allowedTools/systemPrompt options to runClaudeStreaming for non-ambient test scenarios All 6 integration tests pass: Simplifier, Scrutinizer, Reviewer, Coder, Designer, Git — confirming YAML block-list skills are preloaded correctly.
build-hud.js previously copied only dist/hud/ to scripts/hud/, leaving imports to sibling directories (e.g. ../utils/notifications-shape.js) unresolvable at install time. Now walks the ESM import graph from the entry point and copies every reachable file to scripts/ at the matching relative path. Also adds scripts/utils/ to .gitignore.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes skill preload behavior for subagents.