MANIFEST.yaml

# pai-thin audit catalog — v0.1 deliverable
#
# Classifies every meaningful unit of the fresh upstream tree (base 2fde1bb, 2026-05-20)
# against STRATEGY.md categories. Audit-only: no tree files are modified.
#
# Decomposition (size-bound, see STRATEGY.md § v0.1 Definition of Done):
#   - Releases/ (10,488 files) → 1 bulk verdict (seeded below)
#   - Packs/ (1,850 files / 54 packs) → 1 verdict per pack, Ralph loop
#   - top-level + Tools/ + .github/ + images/ → direct
#
# Entry shape: path | unit(file|pack|tree) | verdict(KEEP|REBUILD|DELETE|ADD|FIX)
#              | rationale | cite | confidence(HIGH|MED|LOW)

schema_version: 2
generated_against: "upstream HEAD 2fde1bb (2026-05-20), verified fresh 2026-05-28"
generated_by: "ralph-P1 (regenerated clean after scope lock)"

# Distribution summary — updated as the audit fills in.
summary:
  total_units: 76       # 52 packs + Releases/ tree + 23 top-level/Tools/.github/images
  keep: 47
  rebuild: 10
  delete: 3
  add: 6
  fix: 10
  packs_audited: 52     # of 52 (complete)

units:
  - path: Releases/
    unit: tree
    verdict: DELETE
    rationale: >-
      Frozen historical PAI release snapshots (v2.3 through v5.0.0, plus Pi) — 10,488
      files. pai-thin is a curated thin package that ships only the current system;
      carrying upstream's release archive is the opposite of thin and adds no value to
      a CC-native variant. Historical releases remain available upstream.
    cite: "STRATEGY.md § Anti-scope; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Agents
    unit: pack
    verdict: FIX
    rationale: >-
      CC native subagents and Task cover execution and persisted agent definitions, but
      they do not natively compose task-specific workers from reusable traits, voice/prosody
      mappings, colors, and saved personality profiles; that composition layer is unique PAI
      value. Verdict is FIX rather than KEEP because src/Tools/SpawnAgentWithProfile.ts
      imports a removed AgentProfileLoader, and src/Tools/LoadAgentContext.ts points at
      ~/.claude/Skills/Agents while the installer and docs use ~/.claude/skills/Agents.
      The SpawnParallelAgents workflow is mostly native Task guidance and can be thinned
      later, but the trait/voice composition files remain the load-bearing exception.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/ApertureOscillation
    unit: pack
    verdict: KEEP
    rationale: >-
      ApertureOscillation is already grounded as a native Claude Code skill, and
      its core value is a PAI-specific three-pass scope-framing method rather than
      a wrapper around a CC-native feature. Claude Code has no built-in equivalent
      for holding a tactical target constant while oscillating between local and
      strategic frames to surface design tensions. Per-file exception: src/SKILL.md
      still includes an execution-log snippet that writes to PAI's custom MEMORY
      JSONL path, which should be removed or rebuilt with native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Aphorisms
    unit: pack
    verdict: KEEP
    rationale: >-
      Aphorisms is already packaged as a native Claude Code skill, and its core
      value is a curated quote database plus workflows for thematic matching,
      source-aware addition, thinker research, and newsletter usage tracking.
      Claude Code has no native equivalent for this domain-specific collection or
      recommendation procedure. Per-file exceptions: src/SKILL.md and the workflow
      files still mandate Pulse voice notifications, and src/SKILL.md appends to
      PAI's custom MEMORY execution log; those scaffolding pieces should be deleted
      or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/Apify
    unit: pack
    verdict: KEEP
    rationale: >-
      Apify is already packaged as a native Claude Code skill, and its core value is
      a code-first wrapper around Apify actors for social, business, ecommerce, and
      web scraping with filtering before results enter model context. Claude Code can
      run code and use MCP tools, but it has no native Apify actor catalog or token-
      sparing scraping API equivalent. Per-file exceptions: src/SKILL.md and
      src/Workflows/Update.md still mandate Pulse voice notifications, and
      src/SKILL.md appends to PAI's custom MEMORY execution log; those scaffolding
      pieces should be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Art
    unit: pack
    verdict: KEEP
    rationale: >-
      Art is already packaged as a native Claude Code skill, and its core value is
      a domain-specific visual production system: named workflows, prompt templates,
      reference examples, image API adapters, thumbnail composition, and background
      removal guidance. Claude Code can run shell tools and call external services,
      but it has no native equivalent for this curated art-direction layer. Per-file
      exceptions: src/SKILL.md and workflows still require Pulse voice notifications
      and custom MEMORY logging, several workflows reference legacy PAI tool paths,
      and src/Tools/GeneratePrompt.ts is explicitly deprecated pending rewrite.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/ArXiv
    unit: pack
    verdict: KEEP
    rationale: >-
      ArXiv is already packaged as a native Claude Code skill, and its core value is
      a domain-specific research workflow for arXiv search, latest-paper browsing,
      single-paper deep dives, arXiv Atom parsing, and AlphaXiv enrichment. Claude Code
      can fetch URLs and run shell commands, but it has no native equivalent for this
      curated arXiv query/category mapping and paper triage procedure. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding, and
      src/SKILL.md still references legacy PAI customization and custom MEMORY execution
      logging paths that should be removed or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/AudioEditor
    unit: pack
    verdict: FIX
    rationale: >-
      AudioEditor is already packaged as a native Claude Code skill, and its core
      value is a domain-specific audio cleanup pipeline that combines Whisper
      word timestamps, LLM edit classification, ffmpeg cut execution, preview
      mode, and optional Cleanvoice polish; Claude Code has no native equivalent
      for that end-to-end editing workflow. Verdict is FIX rather than KEEP because
      src/Tools/Analyze.ts imports the legacy PAI/TOOLS/Inference.ts wrapper, which
      is not present in the Packs tree and should be replaced with a CC-native or
      local model-call primitive. Per-file exceptions: src/SKILL.md and
      src/Workflows/Clean.md still mandate Pulse voice notifications, and
      src/SKILL.md appends to PAI's custom MEMORY execution log; those scaffolding
      pieces should be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/BeCreative
    unit: pack
    verdict: KEEP
    rationale: >-
      BeCreative is already packaged as a native Claude Code skill, and its core
      value is a research-backed divergent-ideation method: Verbalized Sampling,
      workflow routing, Tree of Thoughts variants, and VS-Multi synthetic corpus
      expansion. Claude Code can think and write creatively, but it has no native
      equivalent for this curated creativity protocol or corpus-expansion procedure.
      Per-file exceptions: src/SKILL.md and the workflow files still mandate Pulse
      voice notifications, src/SKILL.md and SyntheticDataExpansion write to custom
      PAI MEMORY paths, and TechnicalCreativityGemini3 depends on an external Gemini
      CLI workflow that may need native-tooling review later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/BitterPillEngineering
    unit: pack
    verdict: KEEP
    rationale: >-
      BitterPillEngineering is already packaged as a native Claude Code skill, and
      its core value is a prompt-hygiene audit method for identifying over-prompting,
      cross-file contradiction, redundant rules, vague instructions, and load-path
      waste. Claude Code can follow instructions and read settings files, but it has
      no native equivalent for this curated instruction-audit workflow or its
      Five Questions classification scheme. Per-file exceptions: src/SKILL.md still
      mandates Pulse voice notifications, reads PAI user-customization paths, and
      appends to PAI's custom MEMORY execution log; those scaffolding pieces should
      be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/BrightData
    unit: pack
    verdict: KEEP
    rationale: >-
      BrightData is already packaged as a native Claude Code skill, and its core
      value is a cost-aware progressive scraping and crawling workflow that escalates
      from CC-native WebFetch through curl, browser automation, and Bright Data MCP/API
      only when needed. Claude Code provides WebFetch, Bash, and MCP plumbing, but it
      has no native equivalent for this curated bot-detection bypass, crawl-scope,
      and cost-control procedure. Per-file exceptions: INSTALL.md and VERIFY.md are
      generic pack-install scaffolding, while src/SKILL.md and FourTierScrape.md still
      mandate Pulse voice notifications and src/SKILL.md appends to PAI's custom
      MEMORY execution log; those scaffolding pieces should be deleted or rebuilt on
      native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Browser
    unit: pack
    verdict: FIX
    rationale: >-
      Browser is already packaged as a native Claude Code skill, and its core
      value is a curated agent-browser workflow layer for persistent auth profiles,
      batch headless browser actions, screenshots, recipes, and YAML story validation.
      Claude Code has WebFetch, Bash, and Task agents, but no native equivalent for
      this reusable authenticated browser-automation layer. Verdict is FIX rather
      than KEEP because src/SKILL.md deprecates BrowserAgent and UIReviewer, while
      src/Recipes/SummarizePage.md plus the Automate, ReviewStories, and Update
      workflows still route work through those deprecated agents; INSTALL.md and
      VERIFY.md are generic pack scaffolding, and the Pulse/custom MEMORY hooks should
      be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/ContentAnalysis
    unit: pack
    verdict: KEEP
    rationale: >-
      ContentAnalysis is already packaged as a native Claude Code skill, and its
      core value is a content-adaptive wisdom extraction method with dynamic
      section detection, depth levels, conversational tone rules, and an explicit
      quality gate. Claude Code can fetch, read, and summarize content, but it has
      no native equivalent for this curated extract_wisdom successor workflow.
      Per-file exceptions: INSTALL.md and VERIFY.md are generic pack-install
      scaffolding, while ExtractWisdom still references legacy PAI user
      customization and writing-style paths and suggests fabric for YouTube
      transcripts; those support details should be thinned or reviewed later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/ContextSearch
    unit: pack
    verdict: KEEP
    rationale: >-
      ContextSearch provides a curated prior-work recovery workflow that searches
      Claude Code conversation history, current-project git history, native project
      memory files, and optional PAI session stores before synthesizing usable task
      context. Claude Code exposes the underlying history, memory, file, and shell
      primitives, but it has no native equivalent for this single-command,
      multi-source context search or its standalone/paired usage modes. Per-file
      exceptions: src/commands/context-search.md and src/commands/cs.md are legacy
      slash-command definitions rather than native skill-form packaging, and their
      PAI MEMORY/WORK sources depend on the memory subsystem planned for native
      rebuild.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Council
    unit: pack
    verdict: KEEP
    rationale: >-
      Council is already packaged as a native Claude Code skill, and its core
      value is a curated multi-agent debate method with topic-specific composed
      council members, round-by-round friction, visible transcripts, and synthesis.
      Claude Code provides Task/subagent execution primitives, but it has no native
      equivalent for this collaborative-adversarial deliberation workflow. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding,
      while src/SKILL.md and the workflow files still mandate Pulse voice
      notifications and src/SKILL.md appends to PAI's custom MEMORY execution log;
      those scaffolding pieces should be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/CreateCLI
    unit: pack
    verdict: KEEP
    rationale: >-
      CreateCLI is already packaged as a native Claude Code skill, and its core
      value is a curated TypeScript CLI generation system with tier selection,
      llcli-style templates, extension and migration workflows, and production
      quality gates. Claude Code can write files and run build tools, but it has
      no native equivalent for this reusable CLI-first architecture playbook.
      Per-file exceptions: INSTALL.md and VERIFY.md are generic pack-install
      scaffolding, while src/SKILL.md and workflow files still mandate Pulse voice
      notifications, custom MEMORY logging, and PAI-specific generated paths/config
      conventions that should be deleted or thinned later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/CreateSkill
    unit: pack
    verdict: KEEP
    rationale: >-
      CreateSkill is already packaged as a native Claude Code skill, and its core
      value is a curated lifecycle for creating, validating, testing, improving,
      canonicalizing, and trigger-optimizing skills. Claude Code provides the native
      skill substrate, but it has no native equivalent for this skill-authoring
      quality system or its with-skill versus baseline evaluation workflow. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding,
      src/SKILL.md and workflows still mandate Pulse voice notifications and custom
      MEMORY logging, and CanonicalizeSkill/UpdateSkill reference older PAI skill
      documentation paths that should be thinned or corrected later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Daemon
    unit: pack
    verdict: REBUILD
    rationale: >-
      Daemon's public-profile aggregator and deterministic security filter are useful
      PAI-specific functionality with no direct Claude Code native equivalent. The
      implementation should be rebuilt because DaemonAggregator.ts is grounded in
      PAI's parallel MEMORY/WORK, KNOWLEDGE, TELOS, and PROJECTS substrate that
      pai-thin plans to replace with CC auto-memory topic files and path-scoped
      rules. Per-file exceptions: SecurityFilter.ts and SecurityClassification.md
      are keep-worthy enforcement assets, while INSTALL.md, VERIFY.md, Pulse
      notification text, and custom MEMORY execution logging are generic or legacy
      scaffolding to delete or rebuild on native telemetry.
    cite: "DIVERGENCE.md § Memory (P3, P5); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Delegation
    unit: pack
    verdict: DELETE
    rationale: >-
      Delegation is mostly routing guidance for Claude Code native primitives:
      Task/subagents, worktree isolation, background agents, and TeamCreate-style
      coordinated teams. That makes the standalone pack wrapper overhead rather
      than unique PAI value, even though a few heuristics about when to fan out
      work could be folded into Algorithm or agent guidance later. Per-file
      exceptions: src/SKILL.md also references the PAI Agents skill and writes
      custom MEMORY execution logs, while INSTALL.md and VERIFY.md are generic
      pack-install scaffolding.
    cite: "STRATEGY.md § Design principles; STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Evals
    unit: pack
    verdict: FIX
    rationale: >-
      Evals is unique PAI value: Claude Code has native skills, Task/subagents,
      and telemetry surfaces, but no native pass@k/pass^k agent evaluation
      framework with transcript graders, scenario simulation, suite management,
      and failure-to-task workflows. Verdict is FIX rather than KEEP because the
      model-based graders and PAIAgentAdapter import the legacy PAI/TOOLS/Inference.ts
      wrapper, which is absent from the current pack tree and should be replaced
      with a pai-thin-native model-call primitive. Per-file exceptions: INSTALL.md
      and VERIFY.md are generic pack-install scaffolding, while src/SKILL.md and
      several workflows still mandate Pulse notifications and custom MEMORY logging
      that should be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); DIVERGENCE.md § Divergence categories"
    confidence: HIGH
  - path: Packs/ExtractWisdom
    unit: pack
    verdict: KEEP
    rationale: >-
      ExtractWisdom is an opinionated Claude Code skill for adaptive wisdom
      extraction, with depth levels, dynamic section selection, voice rules,
      spicy-take preservation, and references/rabbit-hole output. Claude Code can
      natively read files, fetch pages, and summarize content, but it has no native
      equivalent for this specific extraction rubric and quality contract. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding,
      while src/SKILL.md still references legacy PAI customization and writing-style
      paths that should be thinned or redirected later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles"
    confidence: MED
  - path: Packs/Fabric
    unit: pack
    verdict: KEEP
    rationale: >-
      Fabric is explicitly called out as KEEP because it is a pattern-library
      wrapper with no native Claude Code equivalent: CC can host skills and read
      files, but it does not provide Fabric's 240+ curated prompt patterns or
      pattern-selection workflows. The pack is already mostly re-grounded on
      native execution by reading Patterns/{name}/system.md directly instead of
      spawning the fabric CLI for ordinary use. Per-file exceptions: src/SKILL.md
      and workflows still include Pulse notifications, custom MEMORY execution
      logging, and ExecutePattern's summarize auto-harvest side effect through
      _HARVEST; UpdatePatterns also keeps the Fabric CLI as an upstream-sync helper.
    cite: "STRATEGY.md § keep / native-rebuild / delete sheet; DIVERGENCE.md § Algorithm & Skills (P3)"
    confidence: HIGH
  - path: Packs/FirstPrinciples
    unit: pack
    verdict: KEEP
    rationale: >-
      FirstPrinciples is already packaged as a native Claude Code skill, and its
      core value is a reusable reasoning method for deconstructing problems,
      challenging soft constraints and assumptions, and reconstructing from hard
      truths. Claude Code can reason and use skills natively, but it has no
      built-in equivalent for this specific first-principles workflow and output
      contract. Per-file exceptions: INSTALL.md and VERIFY.md are generic
      pack-install scaffolding, while src/SKILL.md and workflow files still
      mandate Pulse voice notifications and custom MEMORY execution logging that
      should be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Ideate
    unit: pack
    verdict: KEEP
    rationale: >-
      Ideate is already packaged as a native Claude Code skill, and its core
      value is a PAI-specific multi-cycle evolutionary ideation engine with
      consume, dream, steal, mate, test, evolve, and meta-learn phases. Claude
      Code can run skills and spawn agents, but it has no native equivalent for
      this loop controller, structural-randomness protocol, provenance model, or
      fitness-landscape output. Per-file exceptions: INSTALL.md and VERIFY.md
      are generic pack-install scaffolding, workflow files still mandate Pulse
      notifications and custom MEMORY logging/persistence, and Dream.md points
      to a Daydream.md workflow that is not present in this pack.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Interceptor
    unit: pack
    verdict: KEEP
    rationale: >-
      Interceptor is explicitly listed as a PAI-specific skill to keep because
      Claude Code can run skills and browser-adjacent tools, but it has no
      native equivalent for controlling the user's real Chrome session through
      the Interceptor extension, including authenticated-page verification,
      monitor/replay flows, passive network logs, canvas inspection, and rich
      editor scene graph workflows. This makes it unique PAI value rather than a
      wrapper around a native CC primitive. Per-file exceptions: INSTALL.md and
      VERIFY.md are generic pack-install scaffolding, while src/SKILL.md and
      workflow files still mandate Pulse voice notifications and custom MEMORY
      execution logging that should be deleted or rebuilt on native telemetry
      later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Interview
    unit: pack
    verdict: KEEP
    rationale: >-
      Interview is explicitly listed as a PAI-specific skill to keep because its
      value is a phased conversational review across TELOS, ideal-state,
      preference, current-state, and identity context with scan-driven ordering
      and one-question-at-a-time capture. Claude Code can ask questions, edit
      files, and run native skills, but it has no native equivalent for this PAI
      context-audit protocol or its TELOS-specific review/fill rules. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding,
      and src/SKILL.md still depends on legacy PAI tools/Pulse paths that may
      need thinning later without changing the pack-level verdict.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Investigation
    unit: pack
    verdict: KEEP
    rationale: >-
      Investigation is already shaped as a native Claude Code skill, and its
      core value is a curated OSINT and people-finding system: 279 cataloged
      sources, authorization gates, specialized workflows, parallel research
      patterns, source attribution, and confidence scoring. Claude Code can run
      searches, read files, and launch agents, but it has no native equivalent
      for this structured investigation methodology or ethical source catalog.
      Per-file exceptions: INSTALL.md and VERIFY.md are generic pack-install
      scaffolding, while OSINT and PrivateInvestigator instructions still embed
      Pulse notifications, custom MEMORY paths, and PAI Research/custom-agent
      assumptions that should be thinned or rebuilt later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles"
    confidence: MED
  - path: Packs/IterativeDepth
    unit: pack
    verdict: KEEP
    rationale: >-
      IterativeDepth is already packaged as a native Claude Code skill, and its
      core value is a PAI-specific multi-lens method for surfacing hidden
      requirements, anti-criteria, and ISC refinements before execution. Claude
      Code can reason, use skills, and spawn agents, but it has no native
      equivalent for this structured 2-8 pass lens protocol or its research-
      grounded lens catalog. Per-file exceptions: INSTALL.md and VERIFY.md are
      generic pack-install scaffolding, while src/SKILL.md still references
      legacy PAI customization and custom MEMORY execution logging that should
      be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/ISA
    unit: pack
    verdict: KEEP
    rationale: >-
      ISA is explicitly listed as a PAI-specific skill to keep because its core
      value is the Ideal State Artifact methodology: twelve-section structure,
      tier completeness gates, stable ISC IDs, deterministic Reconcile, and
      Scaffold/Interview/CheckCompleteness/Append/Seed workflows. Claude Code
      natively supports skills, file editing, and questions, but it has no
      native equivalent for this specification-and-verification primitive or
      its ID-stable feature-file merge protocol. Per-file exceptions: INSTALL.md
      and VERIFY.md are generic pack-install scaffolding, while src/SKILL.md
      and workflow files still embed Pulse voice notifications and legacy
      ~/.claude/PAI/MEMORY/WORK paths that should be thinned later without
      changing the pack-level verdict.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Loop
    unit: pack
    verdict: KEEP
    rationale: >-
      Loop is unique PAI value because it turns the PAI Algorithm into repeated
      OBSERVE-to-LEARN cycles with evolving ISC criteria, ISA iteration state,
      cumulative learning, and human review between passes. Claude Code can
      continue iterating, run native skills, and use tasks, but it has no native
      equivalent for this Algorithm/ISA loop controller. Per-file exceptions:
      INSTALL.md and VERIFY.md are generic pack-install scaffolding, while
      src/SKILL.md still uses slash-command wording, Algorithm mode, and a legacy
      PAI/ALGORITHM reference that should be aligned when Algorithm is rebuilt as
      a native skill.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Knowledge
    unit: pack
    verdict: REBUILD
    rationale: >-
      Knowledge's useful idea is a typed, cross-linked knowledge workflow for
      notes, ingestion ripples, contradiction review, and graph navigation, but
      the current pack is grounded in PAI's parallel MEMORY/KNOWLEDGE store and
      custom KnowledgeHarvester, MemoryRetriever, and SessionHarvester tools.
      Claude Code now has native auto-memory and standard file access, so the
      archive substrate, harvesting, mining, and retrieval pieces should be
      rebuilt on native memory/topic-file primitives rather than kept as a
      separate memory system. Per-file exceptions: INSTALL.md and VERIFY.md are
      generic pack-install scaffolding, while src/SKILL.md contains the
      domain-specific typed-linking and contradiction workflows worth carrying
      forward in the rebuild.
    cite: "DIVERGENCE.md § Memory (P3, P5); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/Media
    unit: pack
    verdict: FIX
    rationale: >-
      Media is already shaped as a native Claude Code skill, and its core value
      is a unified visual-production layer: routed Art workflows, image
      generation adapters, thumbnail composition, and Remotion video guidance
      that Claude Code does not provide natively. Verdict is FIX rather than
      KEEP because the Art router and README reference missing BrandWallpaper
      and AdHocYouTubeThumbnail workflows, and many workflow commands still point
      at legacy ~/.claude/skills/art or ~/.claude/PAI/Tools paths instead of the
      Media skill layout. Per-file exceptions: INSTALL.md and VERIFY.md are
      generic pack-install scaffolding, while the mandatory Pulse notifications
      and PAI customization paths should be deleted or rebuilt later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/Migrate
    unit: pack
    verdict: REBUILD
    rationale: >-
      Migrate's useful value is a bulk intake workflow that chunks external
      content, routes it into the PAI taxonomy, requires approval for uncertain
      or sensitive destinations, and preserves provenance; Claude Code has no
      native equivalent for that migration/classification protocol. The
      implementation should be rebuilt around native CC auto-memory topic files,
      path-scoped rules, and standard file tools instead of the parallel
      MEMORY/KNOWLEDGE store, feedback memory files, and release-only
      MigrateScan.ts/MigrateApprove.ts commands. Per-file exceptions: INSTALL.md
      and VERIFY.md are generic pack-install scaffolding, while src/SKILL.md
      carries the workflow worth preserving but also contains the legacy Pulse,
      bun, and ~/.claude/PAI path assumptions.
    cite: "DIVERGENCE.md § Memory (P3, P5); DIVERGENCE.md § CLAUDE.md surgery (P4); STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Optimize
    unit: pack
    verdict: REBUILD
    rationale: >-
      Optimize's useful value is an autonomous hill-climbing protocol for code,
      prompts, skills, and agents using either measurable metrics or LLM-as-judge
      evals; Claude Code can run commands and edit files, but it has no native
      equivalent for that optimization loop. The current pack should be rebuilt
      because src/SKILL.md is framed as a /optimize command that triggers PAI
      Algorithm mode and depends on ~/.claude/PAI/ALGORITHM reference files,
      while pai-thin plans to rebuild Algorithm as an explicit native CC skill
      rather than a mode. Per-file exceptions: README.md, INSTALL.md, and
      VERIFY.md are generic pack scaffolding, while src/SKILL.md carries the
      domain workflow worth preserving.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/PrivateInvestigator
    unit: pack
    verdict: KEEP
    rationale: >-
      PrivateInvestigator is already packaged as a native Claude Code skill, and
      its core value is an ethical public-data people-finding and identity
      verification methodology with source catalogs, reverse lookup procedures,
      public-records guidance, and multi-source confidence scoring. Claude Code
      can search, fetch, and launch Task agents, but it has no native equivalent
      for this domain-specific OSINT workflow or its stop conditions for
      harassment, stalking, and non-public data. Per-file exceptions: INSTALL.md
      and VERIFY.md are generic pack-install scaffolding, while src/SKILL.md and
      the workflows still mandate Pulse notifications and src/SKILL.md appends to
      PAI's custom MEMORY execution log; those scaffolding pieces should be
      deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/PAIUpgrade
    unit: pack
    verdict: REBUILD
    rationale: >-
      PAIUpgrade has unique PAI value as a self-improvement and ecosystem
      monitoring workflow: Claude Code can search, use agents, and run skills,
      but it has no native equivalent for Thread 0 prior-state gating,
      source/reflection mining, and prioritized PAI upgrade synthesis. The
      implementation should be rebuilt because it audits upstream PAI surfaces
      that pai-thin plans to replace, including the parallel MEMORY tree,
      SecurityPipeline hooks, slash-command assumptions, Pulse notifications,
      and PAI Algorithm mode paths. Per-file exceptions: INSTALL.md and
      VERIFY.md are generic pack scaffolding, while src/Tools/Anthropic.ts and
      sources.json carry useful source-monitoring logic that should be
      re-grounded in the native skill layout.
    cite: "DIVERGENCE.md § Memory (P3, P5); DIVERGENCE.md § Hooks (P3); DIVERGENCE.md § Algorithm & Skills (P3)"
    confidence: HIGH
  - path: Packs/Prompting
    unit: pack
    verdict: FIX
    rationale: >-
      Prompting is already shaped as a native Claude Code skill, and Claude Code
      has no built-in equivalent for this reusable prompt-engineering standards
      library, Handlebars prompt-template system, and render/validate tooling for
      generated agent and eval prompts. Verdict is FIX rather than KEEP because
      the root src/Tools renderers resolve template paths as if templates live at
      the skill root while the documented Primitives/... examples only line up
      with src/Templates/Tools, and the pack still includes legacy Pulse
      notifications plus custom PAI MEMORY execution logging. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding,
      while src/Templates/Tools/index.ts is leftover Bun scaffold rather than
      Prompting functionality.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/RedTeam
    unit: pack
    verdict: KEEP
    rationale: >-
      RedTeam is already packaged as a Claude Code skill, and its core value is
      a PAI-specific adversarial-analysis methodology for stress-testing ideas,
      plans, architecture choices, and competing proposals. Claude Code has
      native Task agents and skill loading, but no native equivalent for the
      32-perspective red-team protocol, steelman/counter-argument discipline,
      and adversarial validation workflow. Per-file exceptions: INSTALL.md and
      VERIFY.md are generic pack-install scaffolding, while src/SKILL.md and
      the workflows still contain legacy Pulse notification and custom PAI
      MEMORY execution-log assumptions that should be deleted or rebuilt later.
    cite: "STRATEGY.md § keep / native-rebuild / delete sheet; DIVERGENCE.md § Algorithm & Skills (P3)"
    confidence: HIGH
  - path: Packs/Remotion
    unit: pack
    verdict: KEEP
    rationale: >-
      Remotion is already packaged as a native Claude Code skill, and its core
      value is a curated programmatic-video production workflow: frame-driven
      React animation rules, render tooling, PAI theme integration, AI-video
      pipeline guidance, captions, Lambda rendering, and topic-specific Remotion
      references. Claude Code can edit files and run Bash commands, but it has
      no native equivalent for this domain-specific video composition and
      rendering methodology. Per-file exceptions: INSTALL.md and VERIFY.md are
      generic pack-install scaffolding, while src/SKILL.md and workflows still
      mandate Pulse voice notifications, legacy PAI customization paths, and
      custom MEMORY execution logging that should be deleted or rebuilt on
      native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/RootCauseAnalysis
    unit: pack
    verdict: KEEP
    rationale: >-
      RootCauseAnalysis is already packaged as a native Claude Code skill, and
      its core value is a domain-specific incident-investigation methodology:
      Five Whys, Fishbone/Ishikawa, blameless postmortems, Fault Tree Analysis,
      Kepner-Tregoe IS/IS-NOT, method selection, and foundation references.
      Claude Code can read files, run commands, and invoke skills, but it has
      no native equivalent for this structured RCA decision tree and bias
      controls. Per-file exceptions: INSTALL.md and VERIFY.md are generic
      pack-install scaffolding, while src/SKILL.md still mandates Pulse voice
      notifications, PAI customization paths, and custom MEMORY execution
      logging that should be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Research
    unit: pack
    verdict: KEEP
    rationale: >-
      Research is already packaged as a native Claude Code skill, and its core
      value is a reusable research operating system: depth-based research modes,
      parallel researcher orchestration, mandatory URL verification, iterative
      deep-investigation templates, content retrieval, and extraction workflows.
      Claude Code has native WebSearch/WebFetch and Task agents, but no native
      equivalent for this domain-specific routing, synthesis, verification, and
      investigation-vault methodology. Per-file exceptions: INSTALL.md and
      VERIFY.md are generic pack-install scaffolding, src/SKILL.md requires a
      legacy Pulse notification, and several workflows still write to PAI
      MEMORY/History paths or delegate to the separate Fabric skill and should
      be thinned or rebuilt on native memory/skill primitives later.
    cite: "STRATEGY.md § keep / native-rebuild / delete sheet; DIVERGENCE.md § Algorithm & Skills (P3)"
    confidence: HIGH
  - path: Packs/Sales
    unit: pack
    verdict: FIX
    rationale: >-
      Sales is already packaged as a native Claude Code skill, and its core
      value is a domain-specific workflow for turning product documentation
      into sales narratives, emotional registers, charcoal visual concepts,
      talking points, and script snippets. Claude Code can write pitches and
      invoke tools, but it has no native equivalent for this curated sales-story
      and visual-asset pipeline. Verdict is FIX rather than KEEP because the
      main workflows invoke a standalone StoryExplanation skill that is not
      present in the audited Packs tree and hard-code ~/.claude/skills/art even
      though the Art pack installs as Art; per-file exceptions include generic
      INSTALL.md/VERIFY.md scaffolding plus legacy Pulse notification and custom
      MEMORY execution-log assumptions in src/SKILL.md.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: MED
  - path: Packs/Science
    unit: pack
    verdict: KEEP
    rationale: >-
      Science is already packaged as a native Claude Code skill, and its core
      value is a PAI-specific scientific-method protocol: goal-first framing,
      plural falsifiable hypotheses, experiment design, measurement, analysis,
      iteration, templates, and cross-skill compliance guidance. Claude Code can
      reason, run commands, and invoke skills, but it has no native equivalent
      for this structured evidence-based operating cycle. Per-file exceptions:
      INSTALL.md and VERIFY.md are generic pack-install scaffolding, while
      src/SKILL.md still mandates Pulse voice notifications, legacy PAI
      customization paths, and custom MEMORY execution logging that should be
      deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Scraping
    unit: pack
    verdict: KEEP
    rationale: >-
      Scraping is already packaged as a native Claude Code skill, and its core
      value is a unified scraping workflow that combines progressive URL
      escalation with Apify actor wrappers that filter large datasets before
      they enter model context. Claude Code has WebFetch, Bash, browser
      automation, and MCP plumbing, but it has no native equivalent for this
      curated bot-detection escalation, crawl-cost control, social/platform
      actor catalog, and token-sparing code-first scraping layer. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding,
      while BrightData/SKILL.md, Apify/SKILL.md, Apify/README.md, and
      Apify/INTEGRATION.md still reference Pulse notifications, legacy PAI
      customization paths, or filesystem-mcp paths that should be thinned or
      updated later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Security
    unit: pack
    verdict: FIX
    rationale: >-
      Security is already packaged as native Claude Code skills, and its core
      value is a domain-specific router across recon, web assessment,
      prompt-injection testing, security-news triage, and annual-report threat
      intelligence. Claude Code has native permissions and autoMode hard-deny
      controls for tool safety, but no native equivalent for these security
      assessment methodologies, references, workflows, and helper tools. Verdict
      is FIX rather than KEEP because AnnualReports tools require
      src/AnnualReports/Data/sources.json, which is not present; per-file
      exceptions also include generic INSTALL.md/VERIFY.md scaffolding plus
      legacy Pulse notifications, custom MEMORY paths, and ~/.claude/PAI
      customization assumptions that should be deleted or rebuilt later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Planned divergences (Hooks)"
    confidence: HIGH
  - path: Packs/SystemsThinking
    unit: pack
    verdict: KEEP
    rationale: >-
      SystemsThinking is already packaged as a native Claude Code skill, and
      its core value is a domain-specific structural-analysis methodology:
      iceberg analysis, causal loop diagrams, systems archetypes, Meadows'
      leverage points, and Novak-style concept maps. Claude Code can reason and
      draw diagrams with tool help, but it has no native equivalent for this
      curated systems-thinking workflow and reference corpus. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack-install scaffolding,
      while src/SKILL.md still mandates Pulse voice notifications, legacy
      ~/.claude/PAI customization paths, and custom MEMORY execution logging
      that should be deleted or rebuilt on native primitives later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Telos
    unit: pack
    verdict: REBUILD
    rationale: >-
      Telos preserves useful PAI methodology for life context, project dependency
      analysis, interview extraction, narrative generation, dashboards, and
      consulting-style reports, but its personal TELOS store duplicates context
      that pai-thin should ground in CC native auto-memory and path-scoped rules.
      Claude Code has no native equivalent for the project-analysis workflows or
      the dashboard/report templates, so those should be kept as skill assets while
      the ~/.claude/PAI/USER/TELOS and ~/.claude/context/life/telos assumptions are
      rebuilt. Per-file exceptions: src/Tools/UpdateTelos.ts also appears broken
      against the documented path and imports legacy hook identity code, and
      src/SKILL.md still mandates Pulse voice notifications.
    cite: "DIVERGENCE.md § Memory (P3, P5); DIVERGENCE.md § CLAUDE.md surgery (P4)"
    confidence: HIGH
  - path: Packs/Thinking
    unit: pack
    verdict: KEEP
    rationale: >-
      Thinking is already packaged as native Claude Code skills, and its core
      value is a set of PAI-specific cognitive methodologies: first-principles
      decomposition, iterative-depth lenses, creative ideation, council debate,
      red teaming, world-model threat testing, and scientific investigation.
      Claude Code can reason, use Task agents, and invoke skills, but it has no
      native equivalent for this curated router, reference corpus, and workflow
      suite. Per-file exceptions: INSTALL.md and VERIFY.md are generic pack
      scaffolding, several mode files still mandate Pulse voice notifications
      and legacy PAI customization paths, WorldThreatModelHarness uses the custom
      MEMORY world-model store, and TechnicalCreativityGemini3 plus one Science
      workflow depend on external Gemini/Evals tooling that should be reviewed
      later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § keep / native-rebuild / delete sheet"
    confidence: HIGH
  - path: Packs/USMetrics
    unit: pack
    verdict: FIX
    rationale: >-
      USMetrics is already packaged as a native Claude Code skill, and its core
      value is a domain-specific pipeline for fetching federal economic data,
      maintaining a US-Common-Metrics dataset, calculating multi-horizon trends,
      and generating cross-metric economic analysis. Claude Code can browse,
      call APIs with Bash, and use skills, but it has no native equivalent for
      this curated 68-indicator metric catalog and analysis workflow. Verdict is
      FIX rather than KEEP because the tools rely on hard-coded Substrate paths
      and contain reliability issues such as a malformed EIA gas-price query in
      GenerateAnalysis.ts; per-file exceptions also include generic INSTALL.md
      and VERIFY.md scaffolding plus legacy Pulse voice notification and
      ~/.claude/PAI customization assumptions in src/SKILL.md.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Utilities
    unit: pack
    verdict: REBUILD
    rationale: >-
      Utilities bundles thirteen already-native-style sub-skills behind a manual
      routing skill; the underlying workflows include unique PAI value such as
      Fabric patterns, Evals, Documents/Parser, AudioEditor, Cloudflare, Browser,
      CreateCLI/CreateSkill, and Aphorisms, but Claude Code's native skill
      auto-activation already provides the routing primitive this pack centralizes.
      Rebuild by keeping useful sub-skill assets as independently discoverable
      native skills and dropping or thinning the top-level dispatcher and generic
      install scaffolding. Per-file exceptions: Fabric is explicitly KEEP in the
      strategy, while AudioEditor and Browser inherit the reliability caveats
      already noted in their standalone pack entries; many sub-skills also carry
      legacy Pulse notifications, ~/.claude/PAI customization paths, or custom
      MEMORY assumptions that should be deleted or rebuilt later.
    cite: "DIVERGENCE.md § Algorithm & Skills (P3); STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/Webdesign
    unit: pack
    verdict: FIX
    rationale: >-
      Webdesign has unique PAI value as an orchestration bridge around the
      web-only Claude Design surface through Interceptor; Claude Code's native
      frontend-design plugin can generate frontend code, but it does not
      programmatically drive an authenticated claude.ai/design session or process
      its handoff bundles. Verdict is FIX rather than KEEP because the workflows
      reference tool verbs and verification modes that the bundled tools do not
      implement, including upload/adjust/comment/compare/export tokens and
      lighthouse/compare checks. Per-file exceptions: ProcessHandoffBundle.ts is
      a self-contained KEEP-grade parser, while INSTALL.md and VERIFY.md are
      generic pack scaffolding.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/WorldThreatModel
    unit: pack
    verdict: REBUILD
    rationale: >-
      WorldThreatModel has unique PAI value as a persistent 11-horizon threat-model
      harness for stress-testing ideas, strategies, and investments; Claude Code has
      native skills and Task agents, but no built-in equivalent for this curated
      world-model workflow. The pack should be rebuilt because its core model state
      lives under PAI's custom MEMORY/RESEARCH/WorldModels tree, which pai-thin plans
      to replace with CC-native memory/topic-file patterns. Per-file exceptions:
      INSTALL.md and VERIFY.md are generic pack scaffolding, while src/SKILL.md and
      workflows still mandate Pulse voice notifications and custom MEMORY execution
      logging that should be deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Memory (P3, P5); STRATEGY.md § Design principles (CC first)"
    confidence: HIGH
  - path: Packs/WriteStory
    unit: pack
    verdict: KEEP
    rationale: >-
      WriteStory is already packaged as a native Claude Code skill, and its core
      value is a curated fiction-writing methodology: seven narrative layers,
      Storr character-change mechanics, Pressfield structure, rhetorical figures,
      anti-cliche checks, story-bible construction, and revision workflows. Claude
      Code can write prose and invoke skills, but it has no native equivalent for
      this domain-specific reference corpus and workflow suite. Per-file
      exceptions: INSTALL.md and VERIFY.md are generic pack scaffolding, while
      src/SKILL.md still mandates Pulse voice notifications, legacy PAI
      customization paths, and custom MEMORY execution logging that should be
      deleted or rebuilt on native telemetry later.
    cite: "DIVERGENCE.md § Divergence categories; STRATEGY.md § Design principles (CC first)"
    confidence: HIGH

  # --- Top-level files, Tools/, .github/, images/ (direct audit, in-session 2026-05-28) ---
  - path: README.md
    unit: file
    verdict: ADD
    rationale: "pai-thin rewrote upstream's README into its own differentiated readme (pai-thin vs upstream PAI vs pai-lite). pai-thin-authored content."
    cite: "STRATEGY.md § Fork & first-commit plan"
    confidence: HIGH
  - path: STRATEGY.md
    unit: file
    verdict: ADD
    rationale: "pai-thin-only design/strategy doc with the v0.1 Definition of Done. No upstream equivalent."
    cite: "STRATEGY.md § v0.1 Definition of Done"
    confidence: HIGH
  - path: DIVERGENCE.md
    unit: file
    verdict: ADD
    rationale: "pai-thin-only divergence catalog. No upstream equivalent."
    cite: "DIVERGENCE.md (self)"
    confidence: HIGH
  - path: MANIFEST.yaml
    unit: file
    verdict: ADD
    rationale: "This audit catalog itself — pai-thin v0.1 deliverable. No upstream equivalent."
    cite: "STRATEGY.md § v0.1 Definition of Done"
    confidence: HIGH
  - path: UPSTREAM_README.md
    unit: file
    verdict: ADD
    rationale: "Preserved copy of upstream's README so the original is not lost after pai-thin's README rewrite. pai-thin-only."
    cite: "STRATEGY.md § Fork & first-commit plan"
    confidence: HIGH
  - path: .github/UPSTREAM_SYNC.md
    unit: file
    verdict: ADD
    rationale: "pai-thin-only upstream-pull procedure (PAI has no native upstream-sync). No upstream equivalent."
    cite: "STRATEGY.md § Open questions (upstream-pull mechanism)"
    confidence: HIGH
  - path: LICENSE
    unit: file
    verdict: KEEP
    rationale: "Upstream license, unchanged. Fork retains it."
    cite: "STRATEGY.md § Design principles"
    confidence: HIGH
  - path: SECURITY.md
    unit: file
    verdict: KEEP
    rationale: "Upstream security policy doc; still accurate as policy. No change planned."
    cite: "STRATEGY.md § Design principles"
    confidence: MED
  - path: .gitattributes
    unit: file
    verdict: KEEP
    rationale: "Upstream git attributes, unchanged."
    cite: "STRATEGY.md § Design principles"
    confidence: HIGH
  - path: .gitignore
    unit: file
    verdict: KEEP
    rationale: "Upstream gitignore plus pai-thin's local edits; still correct."
    cite: "STRATEGY.md § Design principles"
    confidence: HIGH
  - path: .env.example
    unit: file
    verdict: KEEP
    rationale: "Env template; harmless scaffold useful to fresh installs."
    cite: "STRATEGY.md § Design principles"
    confidence: MED
  - path: PLATFORM.md
    unit: file
    verdict: KEEP
    rationale: "Upstream platform documentation. Keep, but flag for doc-sync after the hook/memory/security rebuilds land (post-v0.1) since it may describe surfaces that change."
    cite: "STRATEGY.md § REBUILD-ON-NATIVE"
    confidence: MED
  - path: .pai-protected.json
    unit: file
    verdict: REBUILD
    rationale: "Protected-paths config for PAI's custom security pipeline. STRATEGY moves security to CC-native permissions.{allow,ask,deny}; this protection concern is re-grounded there rather than a bespoke JSON consumed by validate-protected.ts."
    cite: "STRATEGY.md § REBUILD-ON-NATIVE (SecurityPipeline); DIVERGENCE.md § Hooks (P3)"
    confidence: MED
  - path: Tools/validate-protected.ts
    unit: file
    verdict: REBUILD
    rationale: "Validator for .pai-protected.json — part of the custom security surface that moves to CC-native permissions. Re-ground with .pai-protected.json."
    cite: "STRATEGY.md § REBUILD-ON-NATIVE (SecurityPipeline)"
    confidence: MED
  - path: Tools/BackupRestore.ts
    unit: file
    verdict: KEEP
    rationale: "Backup/restore utility for the PAI install; no CC-native equivalent. Keep."
    cite: "STRATEGY.md § Design principles"
    confidence: MED
  - path: Tools/README.md
    unit: file
    verdict: KEEP
    rationale: "Docs for Tools/; keep alongside retained tools."
    cite: "STRATEGY.md § Design principles"
    confidence: HIGH
  - path: Tools/utilities-icon.png
    unit: file
    verdict: KEEP
    rationale: "Asset for Tools/ docs."
    cite: "STRATEGY.md § Design principles"
    confidence: HIGH
  - path: .github/FUNDING.yml
    unit: file
    verdict: DELETE
    rationale: "Upstream maintainer's funding config; does not belong in Austin's fork. Low stakes."
    cite: "STRATEGY.md § Anti-scope"
    confidence: MED
  - path: .github/workflows/claude-code-review.yml
    unit: file
    verdict: KEEP
    rationale: "GitHub Actions Claude code-review workflow; useful CI for the fork. May need secret/repo references updated, but the surface is worth keeping."
    cite: "STRATEGY.md § Design principles"
    confidence: MED
  - path: .github/workflows/claude.yml
    unit: file
    verdict: KEEP
    rationale: "GitHub Actions Claude workflow; useful CI. Same caveat as claude-code-review.yml."
    cite: "STRATEGY.md § Design principles"
    confidence: MED
  - path: Packs/README.md
    unit: file
    verdict: KEEP
    rationale: "Index/readme for the Packs/ per-skill layout. Keep."
    cite: "STRATEGY.md § Design principles"
    confidence: HIGH
  - path: Packs/pai-packs-icon.png
    unit: file
    verdict: KEEP
    rationale: "Asset for Packs/ docs."
    cite: "STRATEGY.md § Design principles"
    confidence: HIGH
  - path: images/
    unit: tree
    verdict: KEEP
    rationale: "23 branding/doc image assets. Keep as a tree for v0.1; prune any unreferenced after the README/doc set stabilizes (post-v0.1 execution pass). Low cost, low risk."
    cite: "STRATEGY.md § Design principles"
    confidence: MED