@vicoa/cli 1.9.2 → 1.9.3-win32-x64

package/bin/_internal/integrations/headless/tests/fixtures/omp/README.md

@@ -0,0 +1,108 @@
+# OMP (`oh-my-pi`) RPC wire traces
+
+Real frames captured from `omp --mode rpc`, one JSONL file per scenario, one
+frame per line, verbatim except for the sanitization noted below. These are the
+ground truth for the Pi/OMP wrapper's transport and event mapper — write the
+parsers against these, not against a hand-written idea of the protocol.
+
+Captured **2026-09-05** against:
+
+| | |
+|---|---|
+| binary | `omp` 18.1.10 (`@oh-my-pi/pi-coding-agent`, needs Bun ≥ 1.3.14) |
+| model | `claude-haiku-4-5-20251001` (via `--model haiku`) |
+| flags | `--mode rpc --no-session --model haiku` (+ per-scenario extras) |
+
+Regenerate with `capture.py` in this directory (see "Reproducing" below).
+
+## Scenarios
+
+| file | what it exercises | notable frames |
+|---|---|---|
+| `01-text.jsonl` | plain text turn, no tools | full `thinking_*` → `text_*` delta sequence, `turn_end` |
+| `02-tools.jsonl` | native `read` tool | `tool_execution_start` / `_end` |
+| `03-hosttool.jsonl` | full host-tool round trip | `host_tool_call`, our `host_tool_update` / `host_tool_result` reflected back |
+| `04-approval.jsonl` | `--approval-mode always-ask` | the real permission dialog |
+| `05-todo.jsonl` | todo list creation | `todo_reminder` + the native `todo` tool call |
+| `06-subagent.jsonl` | `set_subagent_subscription: "events"` | `subagent_lifecycle` ×2, `subagent_progress` ×31, `subagent_event` ×110 |
+
+## Sanitization
+
+Only `available_commands_update` frames were modified: the real ones carry the
+operator's full private slash-command and skill list (76 commands, 31 of them
+personal skills). They are replaced with a small synthetic set that preserves
+the shape, including a `subcommands` entry. **Every other frame is verbatim.**
+No credentials, tokens or home-directory paths appear in any file (the capture
+ran in `/tmp`).
+
+## What these traces proved
+
+Seven places where the real 18.1.10 wire disagrees with paseo's schemas
+(`packages/server/src/server/agent/providers/omp/rpc-types.ts`, pinned to a
+16.3.9 floor). Paseo uses `z.discriminatedUnion`, so unmodelled event types fail
+`safeParse` and are silently dropped:
+
+1. **`turn_end`** is a real event carrying the terminal assistant message plus
+   `usage`. Not in paseo's union at all.
+2. **`tool_execution_start.intent`** — a human-readable label
+   (`"Reading sample.txt"`, `"Listing Vicoa sessions with limit 5"`). Unmodelled,
+   and better than anything we would synthesize from `args` for a tool-card
+   header.
+3. **`todo_reminder.attempt` / `.maxAttempts`** — unmodelled.
+4. **`available_commands_update` commands carry `subcommands`** — unmodelled.
+5. **`subagent_event.payload.event` includes `advisor_cost_changed` and
+   `thinking_level_changed`** — event types absent from paseo's union entirely.
+6. **`get_session_stats`** returns `sessionId`, `userMessages`,
+   `assistantMessages`, `toolCalls`, `toolResults`, `totalMessages`,
+   `premiumRequests` and `tokens.reasoning` beyond paseo's shape.
+7. **`get_state.model`** carries `int`, `tps`, `thinking:{mode,efforts,supportsDisplay}`,
+   `identity:{class,family,revision}`, `requiresGlyphTokenization`, `tokenizer`,
+   `supportsComputerUse`, `compat:{…}`.
+
+⇒ **Parse extra-allow. Never use a closed union for OMP event types** — dispatch
+on `type` with an explicit default branch that logs and drops.
+
+## Facts the traces settle
+
+- **Delta order** is `thinking_start` → `thinking_delta`* → `thinking_end` →
+  `text_start` → `text_delta`* → `text_end`, each wrapped in
+  `message_update.assistantMessageEvent`. Thinking blocks also carry a
+  `thinkingSignature` in the final message.
+- **`turn_*` is per model round trip, `agent_*` is per user prompt.** A single
+  tool-using prompt produces `agent_start` → (`turn_start`…`turn_end`) ×2 →
+  `agent_end`. Do not treat `turn_end` as end-of-work.
+- **The `prompt` ack can be `data: null`** — not always `{agentInvoked}`. Do not
+  assume an object.
+- **Host tools render as ordinary tools.** A `host_tool_call` also emits the
+  normal `tool_execution_start` / `_update` / `_end` triple, and our
+  `host_tool_update.partialResult` comes back as
+  `tool_execution_update.partialResult`. No separate rendering path is needed —
+  and streaming progress works for free.
+- **`host_tool_call` uses two distinct ids**: `id` (correlation id; the one that
+  must be echoed in `host_tool_result`) and `toolCallId` (the model's
+  `toolu_…` id, which is what the `tool_execution_*` events key on).
+- **Permission prompts arrive as `extension_ui_request` with `method: "select"`**,
+  a multi-line free-text `title` holding the whole prompt, and
+  `options: ["Approve", "Deny"]`. Maps onto `headless/permission.py`'s existing
+  prompt + options rendering.
+- **`extension_ui_request` also fires with `method: "setWidget"`** in plain
+  `--mode rpc` (not just `rpc-ui`), at session start and end. Unknown methods
+  must be ignored silently.
+- **`subagent_event` is a firehose** — 110 frames for one trivial subagent, since
+  it nests the child's entire event stream. Default
+  `set_subagent_subscription` to `"progress"`, not `"events"`, and derive cards
+  from `subagent_lifecycle` + `subagent_progress`.
+- **`subagent_lifecycle.payload.id` is a human-readable name** (`"DelightedDinosaur"`),
+  not a UUID, and `parentToolCallId` is what links a subagent to its Task card.
+
+## Reproducing
+
+Requires an authenticated `omp` on the machine (`omp` → `/login`). Costs a few
+cents on haiku.
+
+```bash
+cd backend/src/integrations/headless/tests/fixtures/omp
+python3 capture.py            # rewrites the *.jsonl files in place
+```
+
+Re-sanitize `available_commands_update` before committing anything regenerated.

package/bin/_internal/integrations/headless/tests/fixtures/provider_usage/codex_wham_usage_free.json

@@ -0,0 +1,41 @@
+{
+  "user_id": "user-REDACTED",
+  "account_id": "acct-REDACTED",
+  "email": "redacted@example.com",
+  "plan_type": "free",
+  "rate_limit": {
+    "allowed": true,
+    "limit_reached": false,
+    "primary_window": {
+      "used_percent": 0,
+      "limit_window_seconds": 2592000,
+      "reset_after_seconds": 2351904,
+      "reset_at": 1791709519
+    },
+    "secondary_window": null
+  },
+  "code_review_rate_limit": null,
+  "additional_rate_limits": null,
+  "model_usage": {},
+  "credits": {
+    "has_credits": false,
+    "unlimited": false,
+    "overage_limit_reached": false,
+    "balance": null,
+    "approx_local_messages": null,
+    "approx_cloud_messages": null
+  },
+  "spend_control": {
+    "reached": false,
+    "individual_limit": null
+  },
+  "rate_limit_reached_type": null,
+  "promo": {
+    "campaign_id": "REDACTED",
+    "message": "To continue using Codex and get access to GPT-5.3-Codex, start a free trial of Plus today (https://chatgpt.com/explore/plus)"
+  },
+  "rate_limit_reset_credits": {
+    "available_count": 0,
+    "applicable_available_count": 0
+  }
+}

package/bin/_internal/integrations/headless/tests/fixtures/provider_usage/codex_wham_usage_plus.json

@@ -0,0 +1,30 @@
+{
+  "user_id": "user-REDACTED",
+  "account_id": "acct-REDACTED",
+  "email": "redacted@example.com",
+  "plan_type": "plus",
+  "rate_limit": {
+    "allowed": true,
+    "limit_reached": false,
+    "primary_window": {
+      "used_percent": 63.4,
+      "limit_window_seconds": 18000,
+      "reset_after_seconds": 7200,
+      "reset_at": 1789000000
+    },
+    "secondary_window": {
+      "used_percent": 41,
+      "limit_window_seconds": 604800,
+      "reset_after_seconds": 400000,
+      "reset_at": 1789400000
+    }
+  },
+  "credits": {
+    "has_credits": true,
+    "unlimited": false,
+    "overage_limit_reached": false,
+    "balance": "12.5",
+    "approx_local_messages": null,
+    "approx_cloud_messages": null
+  }
+}

package/bin/_internal/integrations/headless/tests/fixtures/provider_usage/copilot_user_free_limited.json

@@ -0,0 +1,67 @@
+{
+  "login": "redacted",
+  "access_type_sku": "free_limited_copilot",
+  "analytics_tracking_id": "REDACTED",
+  "assigned_date": "2025-12-14T13:27:16+08:00",
+  "can_signup_for_limited": false,
+  "chat_enabled": true,
+  "cli_enabled": true,
+  "copilot_plan": "individual",
+  "organization_login_list": [],
+  "organization_list": [],
+  "endpoints": {
+    "api": "https://api.individual.githubcopilot.com",
+    "proxy": "https://proxy.individual.githubcopilot.com"
+  },
+  "can_upgrade_plan": true,
+  "quota_snapshots": {
+    "chat": {
+      "overage_count": 0,
+      "overage_permitted": false,
+      "percent_remaining": 100.0,
+      "quota_id": "REDACTED",
+      "quota_remaining": 200.0,
+      "unlimited": false,
+      "timestamp_utc": "2026-09-14T03:47:16.920Z",
+      "has_quota": true,
+      "quota_reset_at": 0,
+      "token_based_billing": true,
+      "credits_used": 0,
+      "remaining": 200,
+      "entitlement": 200
+    },
+    "completions": {
+      "overage_count": 0,
+      "overage_permitted": false,
+      "percent_remaining": 100.0,
+      "quota_id": "REDACTED",
+      "quota_remaining": 2000.0,
+      "unlimited": false,
+      "timestamp_utc": "2026-09-14T03:47:16.920Z",
+      "has_quota": true,
+      "quota_reset_at": 0,
+      "token_based_billing": true,
+      "credits_used": 0,
+      "remaining": 2000,
+      "entitlement": 2000
+    },
+    "premium_interactions": {
+      "overage_count": 0,
+      "overage_permitted": false,
+      "percent_remaining": 0.0,
+      "quota_id": "REDACTED",
+      "quota_remaining": 0.0,
+      "unlimited": false,
+      "timestamp_utc": "2026-09-14T03:47:16.920Z",
+      "has_quota": false,
+      "quota_reset_at": 0,
+      "token_based_billing": true,
+      "credits_used": 0,
+      "remaining": 0,
+      "entitlement": 0
+    }
+  },
+  "token_based_billing": true,
+  "quota_reset_date": "2026-10-01",
+  "quota_reset_date_utc": "2026-10-01T00:00:00.000Z"
+}

package/bin/_internal/integrations/headless/tests/fixtures/provider_usage/copilot_user_pro.json

@@ -0,0 +1,33 @@
+{
+  "login": "redacted",
+  "access_type_sku": "monthly_subscriber",
+  "copilot_plan": "individual_pro",
+  "quota_snapshots": {
+    "chat": {
+      "percent_remaining": 100.0,
+      "unlimited": true,
+      "has_quota": true,
+      "remaining": 0,
+      "entitlement": 0
+    },
+    "completions": {
+      "percent_remaining": 100.0,
+      "unlimited": true,
+      "has_quota": true,
+      "remaining": 0,
+      "entitlement": 0
+    },
+    "premium_interactions": {
+      "overage_count": 0,
+      "overage_permitted": true,
+      "percent_remaining": 37.5,
+      "quota_remaining": 112.5,
+      "unlimited": false,
+      "has_quota": true,
+      "quota_reset_at": 0,
+      "remaining": 112,
+      "entitlement": 300
+    }
+  },
+  "quota_reset_date": "2026-10-01"
+}