prakhar1114/ai_mime_marketplace/fetch-linkedin-message-threads

Preconditions

Chrome must already be logged into LinkedIn. The skill uses browser-harness through $AI_MIME_BROWSER_HARNESS_BIN and opens a temporary LinkedIn tab, then closes that tab when finished.

Inputs

• last_message_within_days: whole number. Include conversations whose last message timestamp is within this many days from the run time.
• maximum_threads: optional whole number, null, or 0. null, missing, or 0 uses the safe default of 20, matching LinkedIn's observed default conversation page size. Positive values limit returned threads to that number. Values above 100 are rejected to keep LinkedIn access low-volume.

Run

Run with:

./run.sh /path/to/inputs.json

run.sh uses the first available interpreter in this order: skill .venv/bin/python, workflow .venv/bin/python, then required $AI_MIME_PYTHON_PATH.

No third-party Python packages are required. If requirements.txt is added later, use these build or manual repair commands:

"$AI_MIME_UV_PATH" venv .venv --python "$AI_MIME_PYTHON_PATH"
"$AI_MIME_UV_PATH" pip install -r requirements.txt --python .venv/bin/python

These install commands are for skill build or manual repair only. Runtime does not create or repair .venv.

Outputs

The script prints one single-line JSON object:

{"event":"workflow_done","outputs":{"threads":[]}}

outputs.threads is a list of:

• contact_name: LinkedIn participant or conversation name.
• last_message_date: UTC ISO timestamp of the last message activity.
• thread_url: LinkedIn Messaging URL that opens the thread in a new tab.
• new: boolean, true when LinkedIn reports unread/new messages for the conversation.

The output also includes count, last_message_within_days, maximum_threads, pages_fetched, and stopped_because.

Progress logs

Progress logs are written to stderr in clear, short, natural language suitable for an end-user overlay, such as "Opening LinkedIn...", "Reading LinkedIn profile...", and "Reading thread page 1...".

Fallback

Read references/fallback_plan.md if run.sh fails. The fallback keeps the workflow read-only and prioritizes low-volume browser/API access before any manual UI inspection.

ask_llm decision points

None. The skill does not call ask_llm; extraction and filtering are deterministic.

References

• references/fallback_plan.md: manual and UI-agent fallback steps.