team-telnyx/ai / providers/claude/plugins/telnyx-ai/skills/telnyx-meeting-bot
telnyx-meeting-bot
VerifiedUse when an agent must join, observe, react in, transcribe, summarize, or follow up on a Zoom, Google Meet, Microsoft Teams, or Webex meeting with Telnyx Meeting Bot. Handles vague requests, request-specific live polling, name/phrase and semantic triggers, explicitly authorized speak/chat actions, recovery, and all implemented transcript artifact types.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 29.1 KB |
| references/artifact-selection-and-recovery.md | markdown | 6.6 KB |
| references/repeating-semantic-actions.md | markdown | 4.3 KB |
references/artifact-selection-and-recovery.md
View on GitHub# Artifact Selection and Creation Recovery
Use this protocol for final summaries and every manually requested transcript
artifact.
## Observable Implementation Contract
- When `summarize_on_end` is enabled, the finalizer appends
`transcript.completed` and then calls the normal asynchronous
`create_artifact(..., "summary")` path.
- `get_events` exposes each event's `occurred_at` timestamp.
- Artifact responses expose `id`, `type`, `status`, `content`, `prompt`,
`model_provenance`, `failure_reason`, `created_at`, and `updated_at`.
- The public artifact shape has no `automatic`, `origin`, or final-summary marker.
- Artifact lists are newest-first by `created_at`; do not treat list position as
proof of automatic origin.
- Creation is non-idempotent and returns a durable `pending` row before generation
completes.
Do not invent an automatic-artifact marker that the implementation does not
provide.
## Durable State
Persist per session:
```json
{
"transcript_completed_at": null,
"known_manual_artifact_ids": [],
"unreconciled_unknown_manual_creates": [],
"summary_candidate_ids": [],
"selected_summary_artifact_id": null,
"summary_poll_deadline_at": null,
"summary_creation": {
"state": "not_started",
"attempt_count": 0,
"max_pre_send_retries": 2,
"artifact_id": null,
"last_error": null
}
}
```
Creation states are `not_started`, `dispatching`, `accepted`,
`pre_send_failed`, `rejected`, and `outcome_unknown`. Set one fixed deadline
before the first attempt; recovery never extends it.
Every successful manual creation must add the returned ID to
`known_manual_artifact_ids` immediately. Keep the type, stable custom-prompt
hash, and dispatch start time in the corresponding manual request record. If a
manual create becomes `outcome_unknown`, add that request and artifact type to
`unreconciled_unknown_manual_creates`. Do not infer that any unrecognized
same-type artifact is automatic while that uncertainty remains: client and
service clocks need not be comparable, and the unknown create returned no ID.
## Select the End-of-Meeting Summary
1. Drain the final transcript and page `get_events` until `transcript.completed`
appears. Persist its `occurred_at` as `transcript_completed_at`.
2. During the bounded automatic-summary window, repeatedly re-list artifacts.
A likely automatic candidate has `type: "summary"`, an ID absent from
`known_manual_artifact_ids`, and `created_at >= transcript_completed_at`.
3. Persist every candidate ID. Poll all current candidates; do not lock onto the
first pending artifact and ignore summaries that appear or complete later.
4. First identify the candidate with the smallest non-negative
`created_at - transcript_completed_at` delta because the implementation calls
automatic creation immediately after appending `transcript.completed`. Rank
**all statuses** before filtering by status. It must be the unique minimum and
is eligible only when no same-type unknown create remains unreconciled. If two
candidates share the minimum timestamp, origin is ambiguous. Never use
artifact ID, list order, completion order, or client-clock windows as an
origin tie-breaker; use the labeled transcript-grounded fallback instead.
5. Select that unique closest candidate only if it is `completed`. If it is
`pending`, keep re-listing and polling it and every candidate through the fixed
deadline; never skip to a later completed candidate. If the unique closest is
`failed` at the final reconciliation, use the fallback rather than a later
artifact.
6. Immediately before fallback, re-list once more and poll every current
candidate to terminal status within the remaining deadline.
If `transcript.completed` was not observed, `ended_at` is only a weaker lower
bound. Label selection as unconfirmed and never present a summary created before
that bound as the final meeting summary.
Because the public API has no origin marker, an unknown external caller can create
a manual summary after `transcript.completed` that is indistinguishable from the
automatic one. An outcome-unknown create can likewise leave its ID absent from
`known_manual_artifact_ids`; conservatively quarantine automatic-origin
selection for that artifact type until the unknown request is reconciled to an
ID by stronger host evidence. If timing and known IDs do not identify one unique
candidate, say so and prefer the transcript-grounded fallback over claiming an
artifact is the automatic end-of-meeting summary.
## Manual Create State Machine
1. Persist the fixed poll deadline and bounded `max_pre_send_retries` before the
first call.
2. Immediately before invoking REST or MCP, atomically move the request to
`dispatching` and increment `attempt_count`.
3. If the client proves that no request bytes were sent—for example, local
serialization/validation failure, client-queue rejection, or connection/TLS
failure before request transmission—record `pre_send_failed`. The same durable
request may return to `dispatching` and retry within its original deadline and
retry budget because no artifact could have been created.
4. A complete implementation response with `invalid_request`, `invalid_state`,
`not_configured`, or `not_found` is `rejected`; these checks occur before the
artifact row is created. Correct the cause before any bounded retry.
5. Once any request bytes may have been sent, a timeout, disconnect, process
interruption, malformed response, or unknown server error is
`outcome_unknown`. Recovery from either `dispatching` or `outcome_unknown`
must persist an unreconciled same-type unknown-create marker and
re-list/reconcile only; it must not issue another create. Until stronger host
evidence maps that request to a returned ID, every otherwise-unrecognized
same-type artifact is possible manual output, not evidence of automatic
origin.
6. When the response returns an artifact ID, atomically record `accepted`, the
ID, and the manual ID set before polling it.
Never collapse `pre_send_failed` and `outcome_unknown` into one
`creation_attempted` boolean. Only a proven pre-send failure or a confirmed
pre-creation rejection may retry; ambiguous create outcomes never do.
## Polling and Fallback
Poll every accepted/candidate ID until `completed`, `failed`, or the original
fixed deadline. Read `content.text` only from `completed`. Preserve IDs, status,
provenance, and failure reason. Before falling back, perform the final list/poll
reconciliation above.
If no trustworthy completed service artifact is available, deliver the parent
skill's explicitly labeled transcript-grounded fallback and state whether final
transcript completeness was confirmed.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |