# v1.0.0 — Claude Code Skill Spec Alignment + Progressive Disclosure

This release marks the first stable baseline for AI Research Detective as a Claude Code skill/plugin.

Highlights

• Aligns the three core skills with Claude Code / Agent Skills authoring guidance.
• Refactors long SKILL.md files into concise entrypoints with progressive disclosure.
• Adds explicit startup-state handling for empty folders, raw materials, research plans, existing reports, mixed projects, and partial wiki states.
• Strengthens cold-start safety: no unconfirmed CONTEXT/README writes, no old reports treated as primary evidence, and no automatic workflow continuation before user intent is confirmed.
• Narrows tool pre-approvals and removes non-standard path variables.

Skill Authoring Alignment

Updated core skill frontmatter and behavior:

• research-archivist

  • Manual invocation for write-heavy intake workflows.
  • Clear trigger boundary for ingesting, organizing, and updating research materials/wiki.
  • Argument handling for source paths.
  • Reduced tool pre-approval.

• research-detective

  • Clear trigger boundary for analysis, synthesis, report writing, and revision.
  • Argument handling for source paths or analysis questions.
  • Reduced tool pre-approval.
  • Explicit routing between evidence collection, detective analysis, brief answers, full reports, wiki writeback, and information packs.

• research-reviewer

  • Clear trigger boundary for adversarial review, unsupported-claim checks, and evidence validation.
  • Argument handling for report paths.
  • Keeps multi-agent review orchestration inline.
  • Adds a lightweight “report-only” review path when no full project context exists.

Progressive Disclosure Refactor

The three core SKILL.md files are now concise entrypoints instead of long manuals.

New supporting files include:

• skills/research-archivist/workflows/intake_workflow.md
• skills/research-archivist/workflows/incremental_update.md
• skills/research-archivist/workflows/intake_validation.md
• skills/research-archivist/guides/wiki_quality_rules.md
• skills/research-detective/workflows/evidence_collection.md
• skills/research-detective/workflows/detective_analysis.md
• skills/research-detective/workflows/wiki_writeback.md
• skills/research-reviewer/workflows/claim_extraction.md
• skills/research-reviewer/workflows/adversarial_review.md
• skills/research-reviewer/workflows/review_writeback.md
• skills/research-reviewer/guides/reviewer_output_format.md

Cold-start Improvements

Cold start now works as a state classifier and safe initialization flow.

It distinguishes:

• Empty folders
• Raw-material-only projects
• Research design / execution-plan projects
• Existing report or memo projects
• Mixed projects with raw materials and old outputs
• Existing wiki with missing context

It also classifies files into:

• Primary evidence
• Secondary analysis
• Research design
• Project management material

This prevents old reports, decks, and memos from being treated as first-hand research evidence.

Safety Improvements

• Replaces non-standard ${CLAUDE_PLUGIN_ROOT} references with official ${CLAUDE_SKILL_DIR}-based paths.
• Avoids broad Write, Edit, and Bash pre-approvals for auto-invoked skills.
• Requires user confirmation before writing final CONTEXT.md / README.md.
• Requires explicit user intent before moving from cold-start into intake, analysis, or review.
• Keeps reviewer from becoming a rewriting or polishing workflow.

Validation

All test suites pass:

• shared: 18/18
• research-archivist: 12/12
• research-detective: 61/61
• research-reviewer: 17/17

Additional checks performed:

• Markdown link checks across skill docs.
• Stale-pattern scan for unsupported path variables and broad tool pre-approvals.
• Python script compile smoke test.
• Whitespace check via git diff --check.

Stable Baseline

This release is also tagged as:

• stable-skill-v1

Use this tag as the rollback baseline before experimenting with major redesigns.