[hassan1731996]

What does this Ability do?

Family Story Keeper interviews a family member (a grandparent, a parent, or you) about their life, a few questions at a time, and works through the chapters of a life across sessions. Later anyone at home can ask "how did Grandpa meet Grandma?" and hear the answer. With Google linked, it plays the storyteller's own recorded voice from the family's Google Drive. Without Google, it reads back the transcript.

Suggested Trigger Words

• "family stories"
• "story keeper"
• "record a family story"
• "record grandpa's stories" / "interview mum"
• "play a family story"

("tell me a story" is avoided on purpose because Bedtime Stories already uses it.)

Type

• New community Ability
• Improvement to existing Ability
• Bug fix
• Documentation update

External APIs

• No external APIs
• Uses external API(s): Google Drive API v3 through the user's linked Google account (get_token("google")). This is optional and needs no API key. Without Google linked, no audio is recorded and the ability keeps transcripts only.

Testing

• Tested in OpenHome Live Editor
• All exit paths tested (said "stop", "exit", etc.)
• Error scenarios tested (API down, bad input, etc.)

Live Editor (Google linked):

• Interview: consent question, questions and follow-ups, "Shall we keep going?", stories saved.
• Voice: recording captured while the ability listens, uploaded to the family's Google Drive, and played back in the storyteller's own voice.
• Listening and management: hearing a story, queuing a question for the next session, and deleting.
• Exit paths: "stop", "done" and "never mind" at the prompts all exit cleanly.

Probe: a separate throwaway probe ability confirmed the four platform assumptions the design relies on:

1. start_audio_recording() captures audio while user_response() waits.
2. get_token("google") returns a token.
3. That token can create a folder and upload the .wav to Drive.
4. The downloaded clip plays through play_audio() in music mode.

Offline:

• validate_ability.py passes, and flake8 is clean with the CI settings.
• 58 scripted conversation checks pass against a mocked SDK and a mocked Drive.

Error scenarios are covered only in those offline checks, not live:

• Drive request failures and revoked tokens fall back to transcripts.
• Malformed or empty LLM output is handled.
• No-Google mode keeps transcripts only.

Checklist

• Files are in community/family-story-keeper/
• main.py follows SDK pattern (extends MatchingCapability, has register_capability + call)
• README.md included with description, suggested triggers, and setup
• resume_normal_flow() called on every exit path
• No print() — using editor_logging_handler
• No hardcoded API keys — using placeholders
• No blocked imports (redis, user_config)
• No asyncio.sleep() or asyncio.create_task() — using session_tasks
• Error handling on all external calls
• Tested in OpenHome Live Editor

Anything else?

How it differs from existing abilities. Voice Journal, Voice Memory Capture and Private Notes store notes you dictate about yourself. Future Letter Writer saves a message to yourself, and Bedtime Stories invents fiction. Family Story Keeper interviews someone else, follows up on what they say, keeps track of which parts of a life are still untold, and lets family members queue questions for the next session. It is also the first community ability to use the audio recording API to keep and replay a real person's voice.

Review patterns followed from earlier feedback:

• No does_match(); triggers belong in the dashboard.
• State lives in Context Storage, saved with create_key first and the success field checked. Data is cached in memory to cover the visibility delay.
• Exit phrases only match the whole utterance; longer ways of leaving are caught by the LLM.
• The LLM classifies yes/no, and every confirmation allows a change ("no, it's Nani").
• The session stays open after an interview, so follow-up requests are handled by the ability instead of being improvised by the Agent.
• The user's name and family context come from user_profile.md and user_summary.md, which are read only.
• Speech stays short, with no raw dates.

Reuses existing data, read-only. It reads user_profile.md, user_summary.md, memory_mirror_corrections.md and Social Memory's social_memory.json. This lets it suggest who to interview ("Abdul, your grandfather?") and make questions personal. All of it is fenced off as untrusted data in prompts, never spoken verbatim and never logged.

Privacy.

• Every interview starts with a spoken recording notice that needs a yes.
• Recordings stay in the user's own Drive; only Drive file IDs and transcripts are stored in Context Storage.
• Deleting a person removes their Drive folder.
• A short highlights note for the Agent (family_story_keeper.md, under 180 words) can be turned off with "keep the stories private".
• Logs record intents and counts only.

[github-actions]

✅ Community PR Path Check — Passed

All changed files are inside the community/ folder. Looks good!

[github-actions]

🔀 Branch Merge Check

PR direction: add-family-story-keeper → dev

✅ Passed — add-family-story-keeper → dev is a valid merge direction

[github-actions]

✅ Ability Validation Passed

📋 Validating: community/family-story-keeper
  ✅ All checks passed!

[github-actions]

🔍 Lint Results

✅ __init__.py — Empty as expected

Files linted: community/family-story-keeper/main.py

✅ Flake8 — Passed

✅ All checks passed!

[uzair401]

Hello @hassan1731996, thanks for this, really nice ability overall. The interview flow and the way it builds on what the household already shared both feel very natural. I reviewed it closely and pushed a few fixes directly to this branch:

Recording: if a session was interrupted or hit an error while an answer was being recorded, the microphone could stay on after the ability ended. Recording now always stops when the session ends.

Storyteller names: names were matched as a substring, so once "Dad" was saved, saying "Dadi" or "Dada" matched Dad. "Mumtaz" matched "Mum" the same way. A second relative couldn't be added, and "no, it's Dadi" kept switching back to Dad. Names now match on whole words, so "my dad" still finds Dad but similar names stay separate people.

Play another: story numbers restart for each storyteller, so "play another" could skip a different person's story with the same number. Played stories are now tracked per storyteller.

A few smaller things you could look at in a follow-up:

• If nobody answers, the interview takes several prompts to end. Wrapping up after two silent questions would feel better.
• If saved stories can't be loaded for a moment, the ability treats them as empty, and the next save can overwrite them. Better to skip saving when loading failed.
• Please confirm Drive upload and playback on a device, and mention in the README which Google permissions are needed.