v1.9.1

Changed

• The agent uses the ingfah CLI first when ingfah --version reports
  0.1.0 or newer, and the script otherwise. It switches to the script only
  when the CLI cannot reach the API, never after a guard refusal.

v1.9.0

Added

• The agent may use the ingfah CLI 0.1.0 or newer instead of
  scripts/ingfah_api.py when it is installed. SKILL.md maps the script's
  flags to the CLI's, requires a --dry-run preview and the user's yes
  before --yes, and prefers ingfah agents revise and
  ingfah agents publish for agent revisions. The script stays the default.

Removed

• GET /client/text-chats/channel-configs, /providers, and /{id} from
  the allowlist. Use the canonical /client/text-channel-configs routes,
  which return the same data.

v1.8.0

Added

• GET /client/calls/inbound-admission with inbound_admission:read to
  check current inbound call capacity before a SIP transfer.
• GET /client/text-chats/tags with chat_sessions:read to list and search
  customer tags and their customer counts.
• Automation discovery through GET /client/products/{id}/automations/catalog
  and GET /client/automations/trigger-variables with client_products:read.
• Consulting guidance for choosing a setup, explaining tradeoffs, planning
  a pilot, and measuring results from the user's business goal.

Changed

• Refreshed the bundled Thai guide to 73 pages, including callbacks, speech
  previews, text channels, analytics, and team Automations.
• Updated Automation instructions to the current http and do_not_contact
  catalog and Jinja configuration, with guidance for legacy rows.

Fixed

• Reporting guidance now documents API-key bot mute control rather than
  describing it as dashboard-only.

v1.7.0

Added

• GET /client/text-chat-tools (ai_agents:read), the tools a text agent
  may call in a chat, starting with handoff_to_human.
• The text_chat_conversations:write scope and its two routes, which turn the
  AI on or off for one text conversation: PUT /client/chat-conversations/{uuid}
  and PUT /client/text-chats/{provider_type}/channels/{channel_identifier}/conversations/{provider_conversation_id}.
• references/text-chat-handoff.md: what handoff_to_human does, the four
  questions to settle with a client before binding it, how the bot switch
  behaves, and why dashboard test chats cannot show a handoff.
• references/voice-handoff.md: the cold and warm transfer tools compared,
  what only Ingfah's team can configure (numbers, ring timeout, hold and
  no-answer messages), the questions to settle with a client, and how to
  verify a transfer from the call's messages.

Fixed

• The revision guard now checks a text agent's text_chat_tools. Before, a
  prompt-only revision that omitted them silently unbound handoff_to_human.

v1.6.0

Added

• references/multi-agent-teams.md: building a production team of agents
  that hand one call between them, from a four-agent helpline tuned on its
  first two days of real calls. It covers when to split (and giving an
  agent only the tools its job needs), how the platform runs a team (edges,
  handoff tools named transfer_to_<target slug>, who joins the call, tools
  staying per agent, the prompt rendered once at call start), the default
  and seamless handoff modes, the router/specialist line and return paths,
  the "taking over a call in progress" section every receiving agent needs,
  tool-only handoff turns, post-call results and automations across agents,
  the API steps, testing each agent and edge with the promptfoo template, a
  go-live checklist, and what to watch on the first real calls.
• What an API key cannot do for a team: create phone tools, set the handoff
  mode (an API-created team starts in default mode), or choose an agent's
  slug, so prompts naming handoff tools are written after the shells exist.

Changed

• agents-and-teams.md, testing.md, troubleshooting.md, and the
  SKILL.md task table point to the new file; the team-splitting tips moved
  there.

v1.5.1

Fixed

• SKILL.md frontmatter failed to parse as YAML, so the skill could not be
  installed. The unquoted description contained : (in "Thai words: AI
  Agent Team"), which YAML reads as a new key. The value is now quoted.

v1.5.0

Added

• Lessons from tuning live agents, written without client specifics:
  • prompt-authoring.md: branches and fallbacks (clear-but-unexpected answers,
    accept branches, a general no-information line, per-question mismatch
    branches, routing fixes, absolute-reading prohibitions, scoping
    contradictions), content that must be said despite interruptions,
    mishearing (letter/digit codes, no echo, saying the limit aloud, separate
    ladders), persuasion ladders, knowledge-base search (query wording, facts
    that suppress search, the two-search rule, a search check at the end of the
    task), one-line flow examples, the greeting's own rendering, concrete values
    in examples, asking for silence, no in-call summaries, computed numbers in
    say-as, price tables, scoped pronunciation rules, business-hours gates that
    fail closed, and where to place new text
  • tools.md: tools that ask the customer something, lookup keys, failed
    transfers, returning decisions instead of raw values, exception lists in
    descriptions, sending the day and time together to resolve_date, and a
    caveat on model-side days_from_today comparisons
  • agents-and-teams.md: splitting one agent into a team
  • outcome-design.md: the labels that go wrong most, and schema pitfalls
    (integer enums, stored field order, tool results visible to post-call
    steps but not webhooks)
  • postprocessors.md, faq-authoring.md, troubleshooting.md: instruction
    wording, fixing a repeatedly wrong entry, greeting hang-ups, and cut-off
    numbers from malformed say-as tags

Changed

• The knowledge-base query guidance no longer suggests adding a brand name to
  queries; the company's own name measurably lowers retrieval.

• references/testing.md and templates/promptfoo-suite/: how to set up an
  automated conversation test suite for an agent with promptfoo. The template
  exports the agent and its tools through the client API, composes the prompt
  in the platform's order, plays whole conversations on the production model
  (google/gemma-4-26b-a4b-it, temperature 0.5, 768 output tokens), and tests
  the disposition outcomes on google/gemini-3.1-flash-lite. It includes
  reusable assertions (tool fired, silence after a transfer, no tool name
  spoken, spoken-Thai digit matching), a fictional example agent to check the
  setup, token and cost reporting, and guidance on keys, cost, what to test,
  and how to write assertions that survive run-to-run variation. Verified end
  to end: 8 conversations × 3 repeats and 4 disposition cases passing, with
  assertions confirmed to fail when their behaviour breaks.

v1.4.0

Added

• references/prompt-authoring.md:
  • where the template engine (gonja v2.9.0) differs from Python Jinja, as a
    table of traps that fail silently, probed on the production version
  • how to parse {{now}} safely (its current shape, the non-breaking-space
    history, a validity check with a fallback), and a preference for {{hour}}
    and the other dedicated variables
  • computing derived values onto customerContext so the body stays
    cacheable
  • the 1024-character limit on ai_instruction_identity
  • silence after a successful transfer
  • "Scripted lines are copied, so scope them" and "How scripted to make it"
    (controlled variety as the default, with the measured cost of going fully
    free-form)
  • "Testing a change": re-test beyond the edited state, treat greeting edits
    as affecting every call, repeat a scenario before concluding, and a go-live
    checklist of guardrail and complaint cases
• references/tools.md: what resolve_date understands and returns, checked
  against the platform source: the phrase coverage, thai_readback, the two
  meanings of needs_confirmation, past dates, weekend-only is_business_day,
  and how to order the instruction.

v1.3.0

Fixed

• Publishing a prompt edit could strip every tool from a live agent. A
  revision stores only the tool ids it is sent, and
  GET /client/agents/{slug}/revisions/{id} returns none, so a body rebuilt
  from the revision read unbound the transfer and hang-up tools. The agent
  then read transfer_to_human_agent{} aloud instead of transferring and
  could not hang up. scripts/ingfah_api.py now reads the agent before
  POST /client/agents/{slug}/revisions and refuses a body that omits
  phone_tools or ai_plugin_function_ids, drops an id the published agent
  uses, or has a flow state naming a tool that is not bound.
  --allow-tool-drop overrides it when removing a tool is the intended
  change.

Added

• references/agents-and-teams.md: "Carry the tools into every revision",
  covering where the ids are read from, the preview and post-publish checks, and
  publishing by id when others edit the same agent.
• references/troubleshooting.md: the "reads a tool name aloud / promises a
  transfer that never happens / repeats its goodbye" complaint, in check order
  (bindings, the call's tool messages, then the prompt).
• references/prompt-authoring.md: "Tools in the prompt" (silent transfers,
  never quote a forbidden output, keep the tool description and prompt
  consistent), the end_call_keyword tool as a third requirement for
  hanging up, and two pre-publish checks.
• references/tools.md: a tool is callable only while bound, and a tool's
  description is part of every turn.

v1.2.0

Added

• references/troubleshooting.md and a "When the user reports a problem"
  section in SKILL.md: they map what a user sees in the dashboard (a call
  title or summary, an outcome label, metadata, what the agent says, missed
  calls) to the setting that produces it, and describe how to consult before
  proposing a fix.
• references/postprocessors.md now explains what each post-call result type
  writes and where it shows up. It also notes that a summary postprocessor
  sets both the call title and the summary, and that a team without one falls
  back to a global default with no business context.
• Newly allowlisted API-key routes, confirmed against the backend router:
  • analytics (analytics:read): /client/analytics/* and
    /client/text-analytics/*, including the XLSX report download
  • text chats (chat_sessions:read): /client/text-chats/conversations
    and /client/text-chats/chat-sessions-list (plus CSV)
  • text channels (client_products:read): /client/text-channel-configs
    and the provider catalog, read only
  • customer templates (client_outbound_options:write): create, update,
    and delete /client/outbound/options
  • GET /client/outbound/call-data-records, for per-attempt SIP results
  • GET /client/voices and POST /client/agents/{slug}/publish
• references/reporting.md documents the analytics, text chat, and
  call-attempt routes.

Changed

• Creating, editing, and deleting a customer template is no longer listed as
  dashboard-only. Only copying a template still is.