docs(skills): missing_docs audit bookkeeping (surface map + snapshot)

[hongyi-chen]

Companion bookkeeping PR for a missing_docs drift-watch run. It carries every feature_surface_map.md and surface_snapshot.json change from the run, so the sibling feature PR stays focused and the shared files don't conflict. Independently mergeable in any order: every mapped target already exists on main.

Audit inputs: warpdotdev/warp (public client) and warp-server, both audits exiting 0 with no skipped audits and no unaccounted surface items.

Surface map changes

New GA feature flag — AgentCliLaunchModal → ignore list. It's the one-time modal announcing the Warp Agent CLI (app/src/workspace/view/agent_cli_launch_modal/). It shows once, links out to the CLI get-started page, and has no recurring user-facing surface. Same treatment as OzLaunchModal, OpenWarpLaunchModal, OrchestrationLaunchModal, and CodeLaunchModal. The CLI it announces is documented under src/content/docs/cli/.

New API routes — POST /oauth/consent and GET /oauth/consent/info → internal. They back the web consent screen for third-party OAuth clients (router/handlers/public_api/oauth2.go): /consent/info maps a client_id to the display name shown on the screen, and POST /consent records the grant that GET /oauth/authorize redirects to. They're browser-facing plumbing for the already-internal authorize flow and are absent from warp-server's canonical public spec, so they aren't a documentable public API surface.

Map hygiene — Warp Agent CLI slash commands and settings re-pointed at real pages. Sixteen TuiOnly slash commands (/exit, /status, /api-keys, /theme, /vim-mode, and so on) and several CLI-only settings were mapped to internal with comments saying they weren't documented. That stopped being true once the CLI docs landed: they're documented in cli/reference.mdx, cli/index.mdx, and cli/configuration.mdx. Mapping them to those pages makes the map's claims accurate and keeps the audit honest if a page is ever deleted. appearance.zero_state.extrusion_depth and agents.voice.voice_input_hold_key stay internal — neither is documented anywhere, and extrusion_depth is a rendering detail of the built-in start-screen object.

The three new slash commands (/connect-grok, /manage-billing, /upgrade) are documented in the sibling PR and mapped here alongside their peers.

Snapshot

surface_snapshot.json regenerated with --update-snapshot. It picks up the AgentCliLaunchModal flag, the two OAuth consent routes, and the three new slash commands, so the next run diffs against a clean baseline.

Deferred findings

Nothing was silently dropped. The two remaining categories are deliberately routed elsewhere:

3 public API endpoints missing from the published OpenAPI spec (medium)

• GET /agent/artifacts/{artifactUid}/download (agent_artifacts.go)
• GET /agent/run-by-external-reference (agent_webhooks.go)
• POST /agent/runs/{runId}/scores (run_score_reports.go)

All three are present in warp-server's canonical public_api/openapi.yaml without x-internal, so they belong in the published developers/agent-api-openapi.yaml subset. They are not hand-documented here: warp-server is private, and the skill routes released API changes through sync-openapi-spec.

That sync is deferred rather than run in this batch, because sync_openapi.py --mode diff shows the regeneration is much wider than these three paths, and it would publish research-preview Agent Memory schemas:

• Paths added: the 3 above. Path removed: /harness-support/transcript. Operations changed: /agent/artifacts/{artifactUid}, /agent/identities, /agent/runs.
• Schemas added: 16, including AgentMemoryCreateConfig, AgentAutoMemoryCreateConfig, AgentMemoryResponse, AgentAutoMemoryResponse, and MemoryStoreAttachmentResponse. These aren't reachable from a memory endpoint (the memory_stores paths are excluded by policy) — they're pulled in transitively through the memory field on CreateAgentRequest/AgentResponse.
• Schemas modified: 19.

Agent Memory is research preview (its gating flag AIMemories is non-GA, which is why oz memory* and /memory_stores/* are gated:AIMemories in this map). Publishing those schemas would expose it in the public API reference. Per the skill's "when unsure, defer — never expose" guardrail, this needs a deliberate decision before the sync runs.

Recommended next step: either have warp-server mark the memory field and its schemas x-internal in the canonical spec, or extend EXCLUDED_PATHS/schema policy in .agents/skills/sync-openapi-spec/scripts/sync_openapi.py, then run sync-openapi-spec as its own PR. Owners for these handlers did not resolve from warp-server's STAKEHOLDERS/CODEOWNERS, so that PR needs manual reviewer assignment.

30 low-severity terminology findings (owned by style_lint)

Reviewed individually. Most are false positives from the substring matcher:

• agent-mode (6 files) — matches URL slugs (/terminal-and-agent-modes/, /warp-drive/agent-mode-context/) and image filenames (agent-mode-locked-universal-input.png), not prose.
• warp ai / ai command (privacy.mdx, faqs.mdx, terminal/entry) — historical telemetry event names in reference tables (Toggle Warp AI, Opened Warp AI), the live feature name "AI Command Search", and one deliberately historical FAQ heading ("What happened to the old Warp AI chat panel?"). None should be renamed.
• generate command (agents/local-agents/generate.mdx) — the page documents the # Generate feature itself.

Two are genuine but belong to a terminology sweep, not this audit:

• AI credits (12 files) — AGENTS.md says use "credits" without the AI prefix, but the billing docs use "AI credits" as the name of one of three buckets alongside compute credits and platform credits. Renaming it needs a product-terminology decision, applied consistently across billing, enterprise, and error pages.
• Warp Terminal (8 files) — style preference for "Warp" unless distinguishing from Oz.

The missing_docs skill explicitly delegates pure wording issues to style_lint, so these are left for a style_lint run.

Clean categories

CLI commands, settings, docs structure (unlisted pages), stale doc references, map hygiene, and changelog review all produced zero findings this run.

Conversation: https://staging.warp.dev/conversation/52a9f9f7-2e0a-43ae-82c4-84a04c9d4eb1
Run: https://oz.staging.warp.dev/runs/019fdd2a-edce-719d-9dc1-678ca78ec7e0

This PR was generated with Oz.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docs	Ready	Preview	Aug 7, 2026 5:18pm