docs: document completion, custom endpoint, and /team surfaces (+ missing_docs bookkeeping for v0.2026.08.26–v0.2026.09.02)

[warp-factories]

What this feature does

Documents three settings-file keys that the all-settings reference was missing (terminal.input.warp_completions_enabled, terminal.input.native_shell_completions_enabled, agents.custom_endpoints), and carries missing_docs drift-watch bookkeeping through v0.2026.09.02.08.27.stable_01 (previously through v0.2026.08.26.17.59.stable_01).

Summary

• Add the two completion toggles under [terminal.input] and the custom-endpoints object under [agents] in all-settings.mdx. The completion rows state the Native shell completions Preview gate (settings widgets and runtime effect stay off until that flag is GA).
• Surface-map + ledger + release marker + snapshot for releases through v0.2026.09.02.08.27.stable_01.

2026-09-04 drift-watch (v0.2026.09.02.08.27.stable_01)

Release gate advanced from v0.2026.08.26.17.59.stable_01 → v0.2026.09.02.08.27.stable_01. No new feature docs — every 09.02 changelog / oz_updates item failed the worthiness gates or was already covered on this PR.

Bookkeeping only:

• Map GET /agent/conversations/{conversation_id}/screenshots/{screenshot_uid}/download -> internal (x-internal: true upstream; dogfood StoredScreenshots).
• Map POST /factory/{uid}/source/validate -> internal (unreleased /factory namespace).
• Prune dead POST /factory/{uid}/benchmarks/tasks/compose-from-run map entry (route removed from warp-server).
• Allowlist terminal/warpify/ssh-legacy as intentionally unlisted (sidebar comment already says so).
• Append ledger verdicts for #14916, #15691, #15625, #15579, #15642, #15605, #15380, #15574, #15714, #15616, screenshot download, source/validate, compose-from-run removal, StoredScreenshots, ssh-legacy, and staleness bulk.
• check_new_release.py --commit + --update-snapshot.

Deferred findings (Gate 0, unchanged): factory REST (/factory/*, /factory-inbox), preview flags (NativeShellCompletions, HistorySearchRankingV2, ShellWidgetHandoff), dogfood (StoredScreenshots, LrcActivitySignal, agents.warp_agent.other.usage_display_unit / PricingTransparency).

2026-09-04 exhaustive re-run (second pass, same release)

A second, deliberately exhaustive pass over the same release. It re-triaged every finding at every severity across every category and confirmed this PR already covers them: re-running the audit against this branch drops the count from 93 findings to 23, all of which are the known terminology false positives itemized in the ledger.

It also found one gap the earlier passes missed, and the miss is worth understanding rather than just fixing.

/team is GA and ungated (StaticCommand TEAM, SlashCommandSurfaces::TuiOnly, Availability::ALWAYS, no feature_flag), and it was the one TUI-capable command absent from the slash command table on agents/cli/reference.mdx — a table that presents itself as the complete list. Two things hid it:

1. Its -> internal map entry. That is correct for the GUI slash-commands page, where the command genuinely does not exist, but the CLI reference maintains a separate table that internal also suppresses.
2. The repo-wide slash check, documented as a known limitation in SKILL.md. A command counts as covered when any page mentions it.

The 2026-08-28 pass ran a surface-scoped check and concluded "all 32 TUI-capable commands covered under agents/cli/"; the 2026-09-02 pass re-ran it against the GUI surface only. This pass ran it against both surfaces, which is what surfaced /team.

Changes here:

• Remap /team from internal to src/content/docs/agents/cli/reference.mdx, with a NOTE recording that 11 sibling TUI-only entries have the same overstated sentinel.
• Append four ledger rows: the /team Gate 1 pass, the both-surface scoped slash re-verification, the sibling-entry map-accuracy deferral, and the weak-coverage re-run.

The docs change itself ships separately in #684 (one PR per feature). This PR merges independently of it: agents/cli/reference.mdx already exists on main, which is all map hygiene requires of a mapped target.

Snapshot regeneration was a no-op — it was already current on this branch.

Additional deferrals from this pass:

• 11 TUI-only slash entries (/exit, /status, /api-keys, /vim-mode, /theme, /voice, /clear, /logout, /statusline, /reset-statusline, /natural-language-detection) are mapped -> internal but do appear in the CLI reference table. Suppression is identical either way, so no audit outcome changes; left alone rather than churning 11 lines on a PR already in review.
• The slash audit's repo-wide scoping is still unfixed in audit_docs.py. Until it is surface-scoped, a command's doc_covered bucket cannot be read as "documented in the right place." This is the third gap that blind spot has produced (/usage, /set-tab-color, /team).

Earlier on this PR (08.26–09.02)

• Map factory /costs/* routes -> internal (replacing removed /metrics/cost-by-pr-size and /metrics/top-prs).
• Map factory-inbox, learning, linear handoff, issue-tracker issues, slack connection-test -> internal.
• Ignore HistorySearchRankingV2 and ShellWidgetHandoff (preview); track dogfood flags via snapshot.
• Document completion + custom_endpoints settings (Gate 1).

Content design plan

Short form (settings reference update):

• Audience: Users editing Warp settings files who need the TOML keys for completion sources and custom inference endpoints.
• Problem: The keys exist in code (app/src/settings/input.rs, app/src/settings/ai.rs) and the Features UI, but all-settings.mdx omitted them.
• Type: Reference row additions on the existing all-settings page (not a new page).

Test plan

• python3 .agents/skills/missing_docs/scripts/audit_docs.py --warp ../warp --warp-server ../warp-server --diff — after map updates, only known potentially_stale_docs terminology noise remains (23).
• Release marker is v0.2026.09.02.08.27.stable_01.
• Snapshot regenerated from current warp master / warp-server develop checkouts.

Reviewers

/cc @bnavetta @dannyneira @warpdotdev/oss-maintainers

Settings rows: warp client settings owners. Bookkeeping / factory-internal routes: dannyneira fallback (warp-server public_api paths unresolved in STAKEHOLDERS).

[vercel]

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

Project	Deployment	Actions	Updated
docs	Ready	Preview	Sep 5, 2026 4:06pm UTC

[warp-for-oss]

@warp-factories[bot]

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /warp-agent-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[warp-factories]

This PR was generated with Warp.

Comment @warp-agent on this PR to send it follow-up work.

 

[warp-for-oss]

Overview

This PR updates the all-settings reference with completion and custom inference endpoint settings, and records missing-docs drift-watch bookkeeping for the release. The custom endpoint row and internal bookkeeping look consistent, but the completion settings documentation appears to publish dogfood-gated behavior as generally available.

Concerns

• The completion settings rows should either be deferred until the native completions feature is GA or explicitly state the availability gate. The diff records NativeShellCompletions as dogfood, and the runtime forces Warp-only completions when that feature flag is off, ignoring both completion toggles for most ungated users.

Verdict

Found: 0 critical, 1 important, 0 suggestions

Request changes

Comment /warp-agent-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[warp-for-oss]

⚠️ [IMPORTANT] These rows document the completion toggles as generally effective, but this PR also records NativeShellCompletions as dogfood-gated and the runtime forces WarpOnly when that flag is off. Public settings docs should either defer these rows until the feature is GA or state the availability gate so stable users do not try settings that are ignored.