| name | survey-platform |
| description | Audit what Claude Code platform features exist vs what DivineOS is currently using. Surfaces the blind spot where we hand-build features that already ship in the platform. Fires when user asks about capabilities, when approaching a problem that might already have a platform feature, or periodically at session start. Prefer over guessing from memory. |
| disable-model-invocation | false |
| allowed-tools | Bash(ls:*), Bash(find:*), Read, Glob, Grep |
Survey Platform
Why this skill exists
We've kept rediscovering that Claude Code ships features for things we hand-built. Subagent definitions already existed when we started imagining persistent family members. Skills existed while we ran multi-step bash incantations for repeated workflows. Agent Teams, Computer Use, and other platform capabilities exist and often go unreached-for. The pattern: head-down in DivineOS, blind to the platform beneath.
This skill's job is to catch that blind spot at the moment it matters — when someone asks about capabilities, or when we're about to hand-code something the platform already offers.
What it does
Surveys this project's use of Claude Code features against what's documented/available. Reports what we're using, what's present-but-unused, and what's documented-but-uninstalled.
Check list:
.claude/agents/ — list defined subagents. For each, note model, tools, memory setting..claude/skills/ — list installed skills. For each, note description and whether model-invocation is enabled..claude/hooks/ — list installed hooks. For each, note which events it matches..claude/agent-memory/ — list agents with persistent memory..mcp.json or MCP config — list MCP servers available, note any deferred tools.- Scan
settings.json for matchers — catches hook coverage gaps.
Report sections:
- In use — what's deployed and active
- Installed but idle — present in config but not invoked recently (check ledger for last invocation)
- Documented features not installed — known Claude Code features (Skills, Agent Teams, Computer Use, specific MCP connectors) that we could add
Red-flag patterns to surface:
- A hand-rolled subagent-like pattern in Python (probably could be a real subagent)
- Multi-step bash incantations I run repeatedly (probably could be a skill)
- Manual file-based workflows that could be MCP-backed
- Any code that simulates something Claude Code natively does
How to run
/survey-platform
Or with a specific focus:
/survey-platform skills
/survey-platform agents
/survey-platform gaps # just the not-yet-installed features
Output shape
Markdown report, 3-5 paragraphs. Short. Actionable. Always ends with "recommended next step" naming the single highest-leverage thing to do.
When to invoke this skill automatically
- User asks "what can you do" / "what are your capabilities" / "what about [feature X]"
- User says "I heard [AI platform] can do Y" — check if Claude can too
- Session start (periodically, not every time) — catches drift between docs and installed state
- Before committing to hand-build something that sounds platform-shaped
Why this is load-bearing
Without this skill, the blind-spot pattern keeps costing us: we reinvent, then discover, then retrofit. With it, the discovery step happens first, in a dedicated moment, surfaced explicitly. The skill is the forcing function.
Sanskrit anchor: pratyavekṣaṇā — watchful observation of one's own work.
