Pocket AI Transcripts

Read transcripts and summaries from Pocket AI (heypocket.com) recording devices. Use when users want to retrieve, search, or analyze their Pocket recordings, transcripts, summaries, or action items. Triggers on requests involving Pocket device data, conversation transcripts, meeting recordings, or audio note retrieval.

Security Scan

The skill's behavior broadly matches its description (reading Pocket transcripts) but it relies on browser-level token extraction via Chrome remote debugging and the separate 'browser' skill without declaring that dependency, and it stores short‑lived auth tokens on disk — an arrangement that is sensitive and worth caution.

Purpose & Capability

The skill's name, description, and API usage (production.heypocketai.com) align with the included client code. However the SKILL.md and scripts require the separate 'browser' skill and Chrome to be run with a profile and remote debugging flags; those dependencies are not declared in the registry metadata. Asking to extract a Firebase bearer token from the browser is coherent for this purpose but should have been declared.

Instruction Scope

Runtime instructions explicitly tell the user to start Chrome with a profile and use a local browser-eval script to read IndexedDB and extract Firebase tokens. This is narrowly scoped to obtaining an auth token for Pocket, but it necessarily grants the skill access to browser storage (which can contain other credentials). The instructions save tokens to ~/.pocket_token.json. There are no suggestions in SKILL.md to limit the profile used or isolate the browser, nor warnings about the sensitivity of IndexedDB access.

Install Mechanism

There is no remote install/download; the skill is instruction + a local python script. No external archives or runtime downloads are invoked by an install spec. This keeps install-time risk low. The script does invoke subprocesses to call the local 'browser' skill's JS files.

Credentials

No env vars or credentials are declared, which matches registry metadata, but the code requires access to a Chrome user profile and the browser-eval scripts under ~/.factory/skills/browser or ~/.claude/skills/browser. Extracting Firebase tokens from the browser is sensitive and granting remote-debugging access to Chrome may expose other site tokens/cookies if not performed on an isolated profile. The skill writes the token to ~/.pocket_token.json (expires in ~1 hour); presence of this file is additional sensitive state.

Persistence & Privilege

The skill does not request always:true, does not modify other skills, and stores only a short-lived token in the user's home directory. It does not request system-wide persistent privileges.

What to consider before installing

This skill appears to do what it says (access Pocket transcripts) but it extracts a Firebase bearer token from your Chrome profile using remote debugging and a separate 'browser' skill. Before installing: (1) inspect the full reader.py and the browser-eval JS to confirm no unexpected network endpoints or data exfiltration; (2) run Chrome with an isolated/new profile (not your main profile) when extracting tokens; (3) review and, if desired, remove ~/.pocket_token.json after use and revoke sessions via Pocket if you suspect exposure; (4) ensure you trust the source repository (there is no declared homepage and the skill registry metadata omits the browser-skill dependency). If you can obtain a token via an official OAuth flow or the vendor's API, prefer that instead. Additional info (full, un-truncated reader.py and the referenced browser eval.js) would increase confidence in the assessment.

Pocket Transcripts

Read transcripts and summaries from Pocket AI devices via reverse-engineered API.

Quick Reference

Function	Description
get_recordings(days, limit)	List recent recordings
get_recording_full(id)	Get transcript + summary + action items
get_transcript(id)	Get raw transcript text
get_summarization(id)	Get markdown summary
search_recordings(query)	Search by text

Setup (One-Time)

1. Start Chrome with User Profile

~/.factory/skills/browser/start.js --profile
# or
~/.claude/skills/browser/start.js --profile

2. Log into Pocket

Navigate to and log in:

~/.factory/skills/browser/nav.js https://app.heypocket.com

3. Extract Token

python3 scripts/reader.py extract

Token is saved to ~/.pocket_token.json and expires in 1 hour.

Usage

List Recordings

from pathlib import Path
import sys
sys.path.insert(0, str(Path.home() / '.claude/skills/pocket-transcripts/scripts'))
from reader import get_recordings, get_recording_full

recordings = get_recordings(days=30, limit=20)
for r in recordings:
    print(f"{r.recorded_at:%Y-%m-%d} | {r.duration_str} | {r.title}")

Get Full Transcript and Summary

full = get_recording_full(recording_id)

print(f"Transcript ({len(full['transcript'])} chars):")
print(full['transcript'][:500])

print(f"\nSummary (markdown):")
print(full['summary'])

print(f"\nAction Items: {len(full['action_items'])}")
for item in full['action_items']:
    print(f"  - {item}")

Search Recordings

results = search_recordings("meeting", days=90)
for r in results:
    print(f"{r.title} - {r.description[:100]}")

API Details

Base URL: https://production.heypocketai.com/api/v1

Auth: Firebase Bearer token from browser IndexedDB

Key Endpoints:

• GET /recordings - List with pagination, filters
• GET /recordings/{id}?include=all - Full data with transcript/summary

Data Structure:

• Transcript: data.transcription.transcription.text
• Summary: data.summarizations[id].v2.summary.markdown
• Action Items: data.summarizations[id].v2.actionItems.items

Token Refresh

Firebase tokens expire in 1 hour. When expired:

1. Ensure Chrome is running with --profile
2. Confirm logged into app.heypocket.com
3. Re-run: python3 scripts/reader.py extract

Data Model

PocketRecording

• id, title, description
• duration (seconds), duration_str (human readable)
• recorded_at, created_at
• has_transcription, has_summarization
• num_speakers
• latitude, longitude (if location enabled)
• tags (list of strings)

PocketSummarization

• summary (markdown formatted)
• action_items (list)
• transcript (raw text)