Capabilities
Everything Pandora's Box can do is a classified, versioned capability. We use a single classification standard — CLASS-001 — so a capability is always one of four things, never an ambiguous "skill". The shipped catalogue:
The CLASS-001 classification
Tools
LLM-callable functions an agent invokes directly — the verbs of the system. 210 shipped.
Heuristic skills
Prompt-time guidance: learned rules and patterns an agent recalls to do a job better. 27 shipped.
Recipe specs
Declarative playbooks (YAML) describing a multi-step workflow, ready to be packaged. 2 shipped.
Packaged skills
Self-contained, runnable skills (a SKILL.md plus code) following the open Agent Skills standard — portable across Claude Code, Codex and Cursor. 15 shipped.
Capability modules
9 shared modules group related capabilities, activated per agent.
aetheria-visual-plan
Scene and beat planning for animated visual production.
audio-narrate
Broadcast-grade narration and voice synthesis.
deck-builder
Branded, editable presentation decks.
live-stream-vision
Real-time visual understanding of a live stream.
marketing
Campaign drafting and multi-channel post generation.
skill-governance
A shipped capability module.
website-publish
Static-site generation and deployment.
writing-toolkit
Long-form authoring, editing and document assembly.
youtube-publishing
Upload, metadata and publishing automation.
Packaged skills
15 runnable, portable skills (SKILL.md + code) in the open repo.
- brand-sweep — brand-consistency pass across assets
- build_board_pack_from_calendar — board pack from calendar data
- build-active-mode-plan — proactive day plan
- compose-board-pack — assemble a finished board pack
- marketing-campaign-draft — draft a campaign end-to-end
- marketing-channel-post — per-channel post generation
- morning-prep — daily briefing prep
- onboard-new-client — client onboarding workflow
- proactive-triage — surface what needs attention
- session-close-extract — capture decisions at session close
- triage-inbox — inbox triage and prioritisation
- website-deploy — build and deploy a site
- weekly-cost-report — weekly cost roll-up
- weekly-trading-review — weekly review workflow
- youtube-upload-and-monitor — upload and track a video
Governed by design
Capabilities are classified, versioned and activated per agent through a single activation matrix, so an operator turns any capability on or off for any agent. The roadmap adds per-skill signing, machine-readable skill cards and instruction-safety scanning before a skill revision can land — aligning with the open Agent Skills governance pattern.