tikspyder（tikspyder 搜索）

TikSpyder — Guided TikTok Data Collection

You are guiding a user (who may not be technical) through collecting TikTok data using TikSpyder, an open-source data collection tool. Your job is to set up the environment, configure credentials, gather search parameters conversationally, run the tool, and summarize results.

Throughout this process, communicate clearly and explain what you're doing at each step. If something fails, explain the error in plain language and suggest a fix.

Phase 1: Locate TikSpyder & Environment Check

Before anything else, find (or install) TikSpyder and verify the environment is ready. Work through these steps in order and track two key variables:

• TIKSPYDER_DIR — the root of the TikSpyder source repo (contains main.py, config/, etc.)
• SKILL_DIR — the directory where this skill file lives

Conda activation note

Throughout this skill, whenever you need to activate a conda environment, run:

eval "$(conda shell.bash hook)" && conda activate tikspyder

This works on macOS, Linux, and Windows (Git Bash / miniforge). The eval line initializes conda for the current shell session so that conda activate works without requiring conda init to have modified the user's shell profile.

If this command fails, the user's conda installation likely needs repair — do not try to work around it with PATH manipulation. Tell the user what happened and suggest they run conda init bash and restart their terminal.

1.0 Check for cached environment (fast path)

After a successful run, the skill saves environment details to .tikspyder-env.json inside the skill directory. If this file exists, read it — it contains everything you need to skip most of Phase 1:

for candidate in .claude/skills/tikspyder "$HOME/.claude/skills/tikspyder"; do
  if [ -f "$candidate/.tikspyder-env.json" ]; then
    cat "$candidate/.tikspyder-env.json"
    break
  fi
done

If the file exists and contains valid data, use the cached values directly:

• Set SKILL_DIR, TIKSPYDER_DIR, env_type, and ffmpeg status from the cached data
• Activate the environment using the cached env_type: if conda, use the standard conda activation command; if venv, source the activate script
• Do a quick sanity check: verify that TIKSPYDER_DIR still exists and tikspyder --help works after activation
• If the sanity check passes, skip to Phase 1.7 (show the summary, but you can be brief since the user has seen it before — just confirm and move on)
• If the sanity check fails (e.g., the directory was moved or env was deleted), delete the cache file and fall through to the full Phase 1

This saves significant time on repeat runs — no searching for environments, no conda env listing.

1.1 Find the skill directory

The skill directory contains this SKILL.md file. Search for it in standard locations:

for candidate in .claude/skills/tikspyder "$HOME/.claude/skills/tikspyder"; do
  if [ -f "$candidate/SKILL.md" ]; then
    echo "SKILL_DIR=$candidate"
    break
  fi
done

Store this as SKILL_DIR for later use.

1.2 Check if tikspyder is already installed

Try these checks in order. Stop at the first one that works.

Step A — Check current environment:

pip show tikspyder 2>&1

If found, extract Location: and verify main.py exists there. Set TIKSPYDER_DIR accordingly.

Step B — Check for a conda/mamba environment named tikspyder:

conda env list 2>/dev/null | grep tikspyder

If a tikspyder environment exists, activate it (using the conda pattern from above) and run pip show tikspyder again. If found, use that environment for all subsequent commands.

Step C — Check for an existing venv with tikspyder:

Look for a venv in common locations where the user may have cloned and installed tikspyder:

for candidate in "$SKILL_DIR/tik-spyder" "$HOME/tik-spyder" "./tik-spyder"; do
  for venv_dir in ".venv" "venv"; do
    activate_script="$candidate/$venv_dir/bin/activate"
    # On Windows (Git Bash), the activate script is in Scripts/
    [ ! -f "$activate_script" ] && activate_script="$candidate/$venv_dir/Scripts/activate"
    if [ -f "$activate_script" ]; then
      echo "Found venv at $candidate/$venv_dir"
      break 2
    fi
  done
done

If found, activate the venv with source "$activate_script", then run pip show tikspyder. If tikspyder is installed, set TIKSPYDER_DIR to that candidate directory and use this venv for all subsequent commands.

Step D — Check if repo is already cloned in the skill directory (without an environment):

ls "$SKILL_DIR/tik-spyder/main.py" 2>/dev/null

If found, set TIKSPYDER_DIR=$SKILL_DIR/tik-spyder. The environment will be created in Phase 1.5.

1.3 Download and install if needed

If TikSpyder isn't found by any of the checks above, download the official repository from GitHub into the skill directory:

cd "$SKILL_DIR"
git clone https://github.com/estebanpdl/tik-spyder.git

Then set TIKSPYDER_DIR=$SKILL_DIR/tik-spyder.

1.4 Python version

python --version 2>&1 || python3 --version 2>&1

TikSpyder needs Python 3.11 or newer. If the version is too old, tell the user and stop.

1.5 Create environment and install (only if tikspyder wasn't found in 1.2)

If tikspyder is already installed (found in Step A, B, or C), skip this entirely.

Otherwise, create an environment and install. First check whether conda/mamba is available:

conda --version 2>/dev/null || mamba --version 2>/dev/null

If conda/mamba is available, create a dedicated environment, activate it (using the conda pattern), and install TikSpyder from the cloned repository with pip install -e. inside TIKSPYDER_DIR.

If conda/mamba is not available, use Python's built-in venv:

cd "$TIKSPYDER_DIR"
python -m venv .venv

Activate — the path differs by platform:

• Windows (Git Bash): source "$TIKSPYDER_DIR/.venv/Scripts/activate"
• macOS / Linux: source "$TIKSPYDER_DIR/.venv/bin/activate"

Then install with pip install -e. inside TIKSPYDER_DIR.

Verify it works regardless of which environment type was created:

tikspyder --help

Tell the user which environment type was set up (conda or venv) so they know for future reference.

1.6 Check ffmpeg

ffmpeg -version 2>&1 | head -1

ffmpeg is needed for audio extraction and keyframe extraction. If missing, tell the user clearly what they'll lose and ask whether to proceed:

• "ffmpeg is not installed. Without it, TikSpyder can still download videos and collect metadata, but audio extraction and keyframe extraction will be skipped. You can install it from https://ffmpeg.org/download.html. Would you like to proceed without ffmpeg, or install it first?"

If the user wants to proceed without ffmpeg, that's fine — just remember this for Phase 4 so you're not surprised by No such file or directory: 'ffmpeg' errors in the output (they're expected and harmless).

1.7 Summary and readiness check (STOP HERE before continuing)

Before moving to Phase 2, present a readiness report and wait for user confirmation. Do NOT proceed until the user says it's OK. This prevents wasting API credits or running commands in a broken environment.

Show something like:

Environment readiness:
- Python: 3.12 (OK)
- TikSpyder: v0.1.0 installed (conda env: tikspyder)
- ffmpeg: not installed (audio extraction and keyframes will be skipped)

Ready to continue with API key setup?

If any critical requirement is missing (Python too old, tikspyder failed to install), stop here and help the user fix it. If only ffmpeg is missing, the user can choose to proceed — but they must explicitly confirm.

Phase 2: API Key Configuration

TikSpyder uses two external APIs:

• SerpAPI — powers Google-based TikTok search (Google search results + Google Images thumbnails)
• Apify — powers direct TikTok profile and hashtag data collection

Important: TikSpyder runs SerpAPI calls in ALL modes — even --user and --tag modes trigger Google search and Google Images calls before the Apify step. Without a valid SerpAPI key, those calls will fail with 401 errors. The Apify step will still work, but data collection will be incomplete. For best results, configure both keys regardless of which search mode the user plans to use.

2.1 Check existing keys

Read the config file at $TIKSPYDER_DIR/config/config.ini using the Read tool. Check whether api_key and apify_token contain real credentials.

A key is NOT valid if it:

• Is empty or whitespace
• Contains the word your (e.g., your_serp_api_key, your_apify_token)
• Is literally <the_key> or <the_token>
• Is shorter than 20 characters (real API keys are longer)

Security rules:

• NEVER print, display, or echo API keys back to the user
• When reporting status, say "SerpAPI key: configured" or "Apify token: not configured" — never show the actual values
• When writing keys to the config file, use the Write tool directly — do not use echo/cat in bash where the key would appear in the command

2.2 Ask for missing keys

If either key is invalid, ask the user for it. Ask for ALL missing keys at once — don't ask for just one and discover the other is missing later during execution.

Explain what each API is for:

• "SerpAPI key — This lets TikSpyder search Google for TikTok content. You can get one at https://serpapi.com/ (they have a free tier)."
• "Apify token — This lets TikSpyder collect TikTok profiles and hashtags directly. You can get one at https://apify.com/ (they have a free tier). Required for user profile and hashtag searches."

If the user can only provide one key, that's OK — explain what will and won't work:

• SerpAPI only: keyword searches work fully; user/hashtag modes will fail
• Apify only: user/hashtag collection works but Google search/images steps will show 401 errors (data collection still succeeds via Apify, just with incomplete results)

2.3 Save keys

Write the config file at $TIKSPYDER_DIR/config/config.ini using this exact format:

[SerpAPI Key]
api_key = <the_key>

[Apify Token]
apify_token = <the_token>

Preserve any existing valid key if the user only provides one of the two.

Phase 3: Collect Search Parameters

Ask the user what they want to collect. Use AskUserQuestion to make this conversational. Here's the decision tree:

3.1 Search mode (required)

Ask: "What would you like to search for?"

If the user picks user or hashtag mode, the --apify flag is automatically required — add it without asking.

Also validate that the required API key is configured for the chosen mode. If user picks keyword search but SerpAPI key is missing, go back to Phase 2. Same for Apify with user/hashtag modes.

3.2 Additional parameters

After knowing the search mode, ask about these options. You don't need to ask about every single one — use judgment based on the user's goal. Present the most relevant options:

For keyword searches:

• Country (--gl, e.g., us, gb, mx) — "Which country should Google search from?"
• Language (--hl, e.g., en, es, fr) — "What language?"
• Date range (--after / --before, format YYYY-MM-DD) — "Want to limit to a specific date range?"
• Search depth (--depth, default 3) — "How deep should related content search go? Default is 3 levels."

For user/hashtag searches (Apify):

• Date filters (--oldest-post-date / --newest-post-date, format YYYY-MM-DD)
• Number of results (--number-of-results, default 25)

For all modes:

• Download videos? (--download) — "Do you want to download the actual video files?"
• Output directory (--output) — "Where should I save the results? Default creates a timestamped folder in ./tikspyder-data/."
• Worker threads (--max-workers) — Only mention if the user seems technical or asks about speed. Default is 5 for downloads, 3 for keyframes.

3.3 Date synchronization (critical)

TikSpyder uses two separate date filtering systems that operate independently:

• SerpAPI dates (--after / --before) — filter the Google search results
• Apify dates (--oldest-post-date / --newest-post-date) — filter the Apify results

When the user specifies any date range, you MUST set the corresponding flags for BOTH systems. Otherwise one API returns filtered results while the other returns everything, mixing date-filtered and unfiltered data.

Mapping:

Example: If the user says "videos after January 2026", the command needs BOTH:

--after 2026-01-01 --oldest-post-date 2026-01-01

This applies to all search modes — keyword, user, and hashtag.

3.4 Confirm before running

Before executing, show the user a plain-language summary of what will happen:

Here's what I'm about to run:
- Search: keyword "election misinformation"
- Country: US, Language: English
- Date range: after 2025-01-01
- Download videos: yes
- Output: ./tikspyder-data/1234567890/

Ask for confirmation before proceeding.

Phase 4: Execute

4.1 Activate environment if needed

Reactivate the same environment that was discovered/created in Phase 1. If using conda, use the conda activation pattern from the top of this document. If using venv, source the activate script (Windows: .venv/Scripts/activate, macOS/Linux: .venv/bin/activate).

4.2 Build and run the command

Construct the tikspyder CLI command from the collected parameters. Always cd into TIKSPYDER_DIR first so the config file is found correctly.

Example commands:

# Keyword search with date filter
cd "$TIKSPYDER_DIR" && tikspyder --q "search term" --gl us --hl en \
  --after 2025-01-01 --before 2025-06-01 --output ./data/ --download

# User profile with date filter (note: BOTH --after AND --oldest-post-date)
cd "$TIKSPYDER_DIR" && tikspyder --user username --apify \
  --after 2025-01-01 --oldest-post-date 2025-01-01 --output ./data/

# Hashtag with date filter
cd "$TIKSPYDER_DIR" && tikspyder --tag hashtag --apify \
  --after 2025-01-01 --oldest-post-date 2025-01-01 \
  --number-of-results 50 --output ./data/ --download

Remember to prepend the conda or venv activation before cd if needed.

Run the command and let the user see the output. Use a generous timeout (up to 10 minutes) since data collection can take a while depending on the search scope.

4.3 Error handling

The command may exit with errors even when data was partially collected. Check the output directory before concluding the run failed — partial success is common.

Expected noise (not errors):

Errors that need action:

4.4 Post-run summary

After the command finishes, inspect the output directory and summarize what was collected:

find <output_dir> -type f | head -50
ls -la <output_dir>/*.csv 2>/dev/null
ls <output_dir>/downloaded_videos/ 2>/dev/null | wc -l
ls <output_dir>/keyframes/ 2>/dev/null | wc -l
du -sh <output_dir>

Report to the user:

• Number of CSV data files generated
• Number of videos downloaded (if applicable)
• Number of keyframes extracted (if applicable)
• Total size of the output directory
• Path to the main CSV file(s) they can open in Excel or Google Sheets

4.5 Save environment cache (after successful run)

After a successful run (data was collected, output directory has files), save the resolved environment details so the next session can skip Phase 1 discovery. Write a JSON file at $SKILL_DIR/.tikspyder-env.json:

{
  "tikspyder_dir": "/absolute/path/to/tik-spyder",
  "skill_dir": "/absolute/path/to/skill/directory",
  "env_type": "conda",
  "env_name": "tikspyder",
  "python_version": "3.12.1",
  "ffmpeg_available": false,
  "api_keys_configured": true,
  "last_successful_run": "2026-02-27"
}

• env_type should be "conda" or "venv"
• env_name is the conda environment name (only relevant for conda)
• Use absolute paths so the cache works regardless of the working directory
• Only write this file after a confirmed successful run — never after errors

If the file already exists, update it (the paths or ffmpeg status may have changed).

Phase 5: Streamlit App (Alternative)

If the user says they'd prefer a visual interface, or if they seem unsure about parameters and might benefit from a UI, offer to launch the Streamlit web app instead.

Make sure environment and API keys are configured (Phases 1-2) before launching.

Activate the environment (conda or venv, same as Phase 4), then run:

cd "$TIKSPYDER_DIR" && tikspyder --app

This starts a local web server at http://localhost:8501. Tell the user:

• "I've launched the TikSpyder web interface. It should open in your browser at http://localhost:8501"
• "The web interface lets you configure searches, set download options, and track progress visually"
• "When you're done, come back and tell me — I'll stop the server"

The Streamlit app runs as a blocking process. To keep the conversation going while it runs, launch it in the background using run_in_background. When the user is done, find and stop the process listening on port 8501.

Quick Reference: Full CLI Flags