feat: show account limits in CLI and TUI status bars

[grig2277]

Summary

• Shows compact active-account limits in both the classic CLI and Ink TUI status bars.
• Refreshes provider usage only in background workers; session.info, message.complete, and status rendering return cached/stale data immediately.
• Uses single-flight stale-while-revalidate caches keyed to a non-secret fingerprint of the active runtime credential.
• Preserves current Codex credential-resolution safeguards and derives ChatGPT-Account-Id only from the already-selected live token.
• Displays a credential label only when the active token exactly matches its credential-pool entry.
• Budgets the Ink account-limit HUD as one optional tail segment, so status/model/context remain pinned and quota text is omitted whole on narrow terminals.

Example: Codex main 5h 83% • weekly 37%.

Review fixes

Addresses the July 12 review:

• No synchronous provider usage fetches remain in TUI hot paths.
• Cold misses and stale entries schedule at most one background refresh per credential key.
• Failures preserve stale data and retry after a short backoff.
• Credential rotation cannot reuse another account's cached payload or label.
• AccountLimitHud participates in the existing essentialWidth / fits() width budget.
• Ink regression coverage includes a long credential label, multiple windows, wide rendering, and whole-segment omission on a narrow terminal.

PR consolidation

This remains the combined CLI + TUI account-limit status PR and supersedes the earlier TUI-only draft #16644.

Verification

Rebuilt from current main at 7b5ba2054721dde998ed47fd4a0f031955278e99; replacement commit: 1b3d9b34c872908fab6d61bec85af3b5b7df9420.

• Python targeted regression: 228 passed.
• Ink targeted render/event regression: 102 passed.
• TypeScript typecheck: passed.
• TUI production build: passed.
• Ruff: passed.
• ESLint: passed.
• git diff --check: passed.

CI status

GitHub created CI run 29208048955, but it is currently action_required: a repository maintainer/admin must approve the fork workflow before jobs can start.

Broader local-suite caveats are environment/baseline-only: the isolated gateway suite lacks a Chromium binary, while unrelated unchanged Ink tests include SSH-environment assertions and one existing timeout. All changed and newly added tests pass.

Closes #16643

[grig2277]

Added follow-up fix for account-limit status binding.

What changed:

• Runtime provider now carries selected credential_id / credential_label forward.
• CLI/TUI agents pin that selected credential metadata.
• CLI/TUI fallback labels from credential pool only apply when the pool entry token matches the active agent.api_key.
• Credential rotation updates pinned credential metadata too.

Local verification:

• pytest tests/hermes_cli/test_runtime_provider_resolution.py tests/tui_gateway/test_make_agent_provider.py tests/test_tui_gateway_server.py tests/test_account_usage.py tests/cli/test_cli_status_bar.py -q → 214 passed
• npm --prefix ui-tui run type-check → passed
• npm --prefix ui-tui run test -- src/__tests__/createGatewayEventHandler.test.ts → 23 passed
• py_compile / git diff --check → passed
• Live TUI check showed Codex main 5h ... • weekly ...

[grig2277]

Rebased on latest main and added a small regression-test fix so TUI provider-resolution tests are isolated from local HERMES_MODEL / HERMES_TUI_PROVIDER env overrides.

Latest local verification:

• py_compile cli.py agent/account_usage.py hermes_cli/runtime_provider.py tui_gateway/server.py run_agent.py gateway/run.py — passed.
• Targeted Python: tests/test_account_usage.py tests/cli/test_cli_status_bar.py tests/test_tui_gateway_server.py tests/hermes_cli/test_runtime_provider_resolution.py tests/tui_gateway/test_make_agent_provider.py — 232 passed.
• Gateway/tool regressions — 6 passed.
• npm run type-check — passed.
• npm run test -- src/__tests__/createGatewayEventHandler.test.ts — 23 passed.
• Full TUI suite — 389 passed.
• npm exec eslint ... — passed.
• npm run build — passed.

[grig2277]

Rebased this PR on the latest main and refreshed the fork branch.

Current head: f6aa3b2a0b7423fc0d0c9b1e33133a2ee478006c

Local verification after the rebase:

• py_compile cli.py agent/account_usage.py hermes_cli/runtime_provider.py tui_gateway/server.py run_agent.py gateway/run.py — passed
• npm --prefix ui-tui run build — passed
• The rebuilt TUI bundle includes the account-limit HUD/status-line code
• Local checkout is up to date with origin/main

GitHub Actions currently appear blocked for this fork PR: the check suites have conclusion=action_required, so no check runs are reported until a maintainer/admin approves workflows.

Could a maintainer approve the workflow run when convenient?

Also, this PR touches TUI/gateway paths in addition to CLI:

• tui_gateway/server.py
• ui-tui/src/...

The linked issue already has comp/tui and comp/gateway; adding those labels here too may help route review. If possible, I would also appreciate reconsidering the priority from P3: this is currently a recurring maintenance pain because daily updates drop the local status-line patch until the PR is merged.

[grig2277]

Refreshed this PR again on current main and fixed a likely post-approval CI blocker.

Current state:

• Base: 436672de0
• Head: 214501b4b
• Compare: behind_by=0, ahead_by=7

What changed in the latest push:

• Rebased the branch onto latest main.
• Added "grig2277@mail.ru": "grig2277" to scripts/release.py AUTHOR_MAP so the repo's Contributor Attribution Check should not fail after workflows are approved.

Latest local verification:

• py_compile cli.py agent/account_usage.py hermes_cli/runtime_provider.py tui_gateway/server.py run_agent.py gateway/run.py — passed.
• npm --prefix ui-tui run build — passed.
• npm --prefix ui-tui run type-check — passed.
• npm --prefix ui-tui test -- src/__tests__/createGatewayEventHandler.test.ts — 38 passed.
• pytest tests/cli/test_cli_status_bar.py tests/test_account_usage.py tests/tui_gateway/test_make_agent_provider.py tests/test_tui_gateway_server.py -k 'not test_session_create_drops_pending_title_on_valueerror' — 208 passed.
• py_compile scripts/release.py — passed.
• Local Contributor Attribution Check logic for grig2277@mail.ru — passed.

Known baseline note:

• tests/test_tui_gateway_server.py::test_session_create_drops_pending_title_on_valueerror still fails on the full targeted set.
• I reproduced the same failure on clean origin/main, so it appears to be an upstream baseline failure, not caused by this PR.

Could a maintainer/admin please approve the fork workflow runs when convenient? They are currently all stopped with action_required.

Also, this PR touches TUI/gateway paths in addition to CLI (tui_gateway/server.py, ui-tui/src/...). The linked issue already has comp/tui and comp/gateway; adding those labels to the PR too may help route review. If maintainers would prefer this split into smaller backend/CLI/TUI PRs instead of one combined PR, I’m happy to reshape it.

[grig2277]

Rebased and force-pushed this PR onto current main.

Current state:

• compare is clean: behind_by=0, ahead_by=7
• PR is mergeable
• GitHub Actions are still blocked on fork workflow approval (action_required)
• local targeted verification is reflected in the PR body

Could a maintainer/admin please approve the workflow runs when convenient?

Also, this PR touches TUI/gateway paths in addition to CLI (tui_gateway/server.py, ui-tui/src/...). If possible, adding comp/tui and comp/gateway labels here too may help route review. If maintainers would prefer this split into smaller CLI/gateway/TUI PRs, I’m happy to reshape it.

[teknium1]

Thanks for combining the CLI and TUI work. The feature still addresses a gap: current main renders account limits through /usage (cli.py:9681-9702), while the classic status bar (cli.py:5055-5119) and Ink StatusRule do not display them.

Problems

• The proposed TUI cache miss fetch is synchronous, yet the PR invokes it from both session.info and message.complete. A provider timeout would delay those hot paths rather than fail open.
• AccountLimitHud bypasses the established essentialWidth/fits() progressive-disclosure budget in ui-tui/src/components/appChrome.tsx:466-533; narrow terminals can clip this new multi-window segment. The added tests cover gateway state, not narrow-width rendering.

Suggested changes

• Refresh TUI limits in a background single-flight path and emit cached/stale data only from hot RPC/response paths.
• Budget the HUD as a progressive segment and add a narrow-width render test.
• Preserve the current Codex credential-resolution safeguards in agent/account_usage.py:439-494 when resolving the stale branch conflict.

Automated hermes-sweeper review.

[teknium1]

Please route this HUD through the same width budget as the other tail segments. essentialWidth/fits() above deliberately preserve the model and context on narrow terminals; this unbudgeted multi-window string can instead be clipped or displace those higher-priority fields. Add a narrow-width StatusRule render test with a long credential label and multiple windows.

[grig2277]

Addressed the July 12 review in 1b3d9b34c872908fab6d61bec85af3b5b7df9420 and rebuilt the PR from current main.

Key fixes:

• TUI hot paths are cache-only; provider usage refresh is background, stale-while-revalidate, and single-flight per active credential.
• Cache identity and optional labels are bound to the active runtime token without storing token material in cache keys.
• Ink quota rendering is one width-budgeted tail segment; narrow terminals keep status/model/context and omit the HUD whole.
• Added regression coverage for cold miss, stale refresh, failures, credential rotation, long labels, multiple windows, and narrow/wide rendering.

Verification: 228 targeted Python tests and 102 targeted Ink tests passed; typecheck, build, Ruff, ESLint, and git diff --check also passed.

@teknium1, please re-review when convenient.