@talqing/mcp 0.15.0 → 0.16.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/SKILL.md +14 -7
- package/package.json +1 -1
package/SKILL.md
CHANGED
|
@@ -89,9 +89,10 @@ An agent is a single object, `config`, plus what you attach to it:
|
|
|
89
89
|
the request that starts a session supplies values that override the declared
|
|
90
90
|
defaults. See **`vars`** below.
|
|
91
91
|
- **`llm`, `stt`, `tts`** — `{provider, model}` from the provider catalog. Text
|
|
92
|
-
agents have
|
|
92
|
+
agents have `llm`, plus an optional `stt` that transcribes voice notes (see
|
|
93
|
+
**Images and files**); voice and video need all three.
|
|
93
94
|
- **`language`** — the one language the caller and the agent speak, shared by
|
|
94
|
-
every model. `null` is Auto.
|
|
95
|
+
every model. `null` is Auto. On a text agent, only with `stt` set.
|
|
95
96
|
- **`timezone`** — an IANA name such as `Asia/Kolkata`. What the three
|
|
96
97
|
`{{system_vars.…}}` clock variables resolve against, on every channel. `null`
|
|
97
98
|
until one of them is used, and then required. Set it whenever the agent has to
|
|
@@ -362,8 +363,9 @@ support the channel, a language the model does not offer, a speed outside the
|
|
|
362
363
|
model's range, an unpublished tool — all are refused with an explanation. Read
|
|
363
364
|
the error and correct it; that is faster than guessing.
|
|
364
365
|
|
|
365
|
-
Some fields are normalized for you. Switching to `text` clears `
|
|
366
|
-
`greeting`, `turn_handling` and `avatar
|
|
366
|
+
Some fields are normalized for you. Switching to `text` clears `tts`,
|
|
367
|
+
`greeting`, `turn_handling` and `avatar`, and keeps `stt` (voice notes on) —
|
|
368
|
+
set it to `null` in the same write unless the user wants voice notes. Switching to `video` fills in an
|
|
367
369
|
`avatar`, and the call ignores `resume_false_interruption` — the avatar cannot
|
|
368
370
|
resume a sentence it has already stopped rendering. Keypad input is voice-only
|
|
369
371
|
and a text agent cannot watch a screen share, so enabling either off its channel
|
|
@@ -380,9 +382,14 @@ voice. Read it before choosing; do not assume a model exists.
|
|
|
380
382
|
|
|
381
383
|
People can send files on WhatsApp (Twilio senders) and in a web chat, and a photo
|
|
382
384
|
on a web voice or video call. Every file is stored on the conversation; photos,
|
|
383
|
-
PDFs and Word (`.docx`) files are also read by the agent,
|
|
384
|
-
|
|
385
|
-
|
|
385
|
+
PDFs and Word (`.docx`) files are also read by the agent, a shared contact card
|
|
386
|
+
is read as its name, numbers and emails, and anything else (videos,
|
|
387
|
+
spreadsheets) is named to it as something it cannot open.
|
|
388
|
+
**Voice notes** are heard only by a text agent with `stt` set (off by default):
|
|
389
|
+
each note is transcribed in one request on that model, so only entries with
|
|
390
|
+
`file_transcription` set qualify. Turn it on when the user's customers send voice
|
|
391
|
+
notes — on WhatsApp in India, nearly always. A call never transcribes one.
|
|
392
|
+
**There is no setting for images.** Whether the agent sees a photo or a scanned
|
|
386
393
|
PDF page is decided by the model it runs: `vision` on its catalog entry. Most
|
|
387
394
|
entries read images; `gemini-3.8-flash`, `gemini-3.7-flash`, `openai/gpt-live-1`
|
|
388
395
|
and the two `grok-voice-*` realtime entries do not — such an agent is told it
|
package/package.json
CHANGED