OOpenclawnet
  • Introduction
  • Docs
  • Scripts
  • Sessions
    • RELEASE CHECKLIST
    • Session 2 guide es
    • Session 2 guide
    • Session 3 guide es
    • Session 3 guide
      • ARQUITECTURA
      • GUIA DESARROLLADOR
      • Introduction es
      • Introduction
      • RELEASE
      • Copilot prompts
      • Slides es
      • Slides
      • Speaker script
    • Session 4 guide es
    • Session 4 guide
    • Session 5 guide es
    • Session 5 guide
    • Speakers
  • Src
  • Tests
Powered by Docsbook
Sessions/Session 3/Slides
Slides esPreviousSpeaker scriptNext

On this page

  • Session 3 — Skills, Storage & Memory
  • Where Sessions 1–2 left off
  • Today's scope, in one slide
  • The session-3 mental model
  • Behavior
  • Boundaries
  • Why we needed extra tools
  • Ollama Monitor — what it is
  • Ollama Monitor — features
  • Color codes
  • Toast triggers
  • Ollama Monitor — install
  • Ollama Monitor — demo
  • Aspire Monitor — what it is
  • Aspire Monitor — features
  • What it watches
  • What it surfaces
  • Aspire Monitor — install
  • Aspire Monitor — demo
  • Why touch tool-calling at all?
  • OpenAI format alignment
  • What changed since Session 2
  • FileSystemTool refactor
  • Tool sanitizers — the new contract
  • Three sanitizers ship with v1
  • PathSanitizer
  • UrlSanitizer
  • JsonArgumentSanitizer
  • End-to-end: what a tool call looks like now
  • A concrete diff: read operation
  • Per-tool metadata is unchanged
  • NDJSON event additions
  • What it buys you
  • What is a skill?
  • Why skills (vs. tools)?
  • The headline bug we fixed
  • The fix: one loader, one source of truth
  • 3-layer storage
  • Why shared storage, not per-agent?
  • agentskills.io frontmatter
  • What gets dropped from old frontmatter
  • FileSystemWatcher hot-reload
  • Snapshot per request
  • Per-agent enablement: enabled.json
  • Default = disabled, fail-closed
  • Skills.razor UI — top-level nav
  • Browse
  • Act
  • Per-agent assignment modal
  • 4-step import wizard
  • v1 import: awesome-copilot only
  • Manual authoring path
  • SKILL.md anatomy in detail
  • Bounded resources (S-11)
  • Hardening invariants S-1..S-12
  • Hardening invariants S-1..S-12 (cont.)
  • Structured logging — 14 events
  • What we DON'T log
  • E2E tests — what passes
  • Implementation waves
  • Bruno's question
  • What was wrong
  • The new defaults
  • Configuration: three sources, one winner
  • OPENCLAWNET_STORAGE_ROOT — one name only
  • ISafePathResolver — one resolver, one rule
  • H-1: storage-root containment, fail-closed
  • H-2: one sanitizer, one resolver
  • H-3: no reparse-point escapes
  • H-4: boundary-safe containment
  • H-5: strict name allowlist
  • H-6: per-agent scoping seam
  • H-7: ACL hardening on credential subdirs
  • H-8: audit every write
  • All eight, side by side
  • Wiring it up
  • Settings UI
  • Migration story
  • The context window problem
  • Summarization strategy
  • SessionSummary entity
  • IMemoryService shape
  • Local embeddings — no API calls
  • Semantic search across past sessions
  • Transparent memory dashboard
  • What's coming next (Session 4 preview)
  • Demo 1 — aspire start walkthrough
  • Demo 2 — curl /api/skills
  • Demo 3 — manual skill drop-in
  • Key insights
  • What we built today ✅
  • Session 4 preview

Was this page helpful?