gdelt-mentions-fetch（gdelt mentions fetch 命令行）

GDELT Mentions Fetch

Core Goal

• Fetch GDELT 2.0 Global Mentions table exports (*.mentions.CSV.zip) from official public endpoints.
• Resolve latest available snapshot via lastupdate.txt.
• Resolve historical snapshots in a UTC range via masterfilelist.txt.
• Persist downloaded files and return machine-readable JSON manifest.
• Keep runtime observable with structured logs and optional log file.

Required Environment

• Configure runtime by environment variables (see references/env.md).
• Start from assets/config.example.env.
• Load env values before running commands:

set -a
source assets/config.example.env
set +a

Workflow

1. Validate effective configuration.

python3 scripts/gdelt_mentions_fetch.py check-config --pretty

1. Inspect the latest available mentions snapshot.

python3 scripts/gdelt_mentions_fetch.py resolve-latest --pretty

1. Dry-run a historical range selection before downloading.

python3 scripts/gdelt_mentions_fetch.py fetch \
  --mode range \
  --start-datetime 20260301000000 \
  --end-datetime 20260301120000 \
  --max-files 3 \
  --dry-run \
  --pretty

1. Fetch files with transport and structure validation.

python3 scripts/gdelt_mentions_fetch.py fetch \
  --mode latest \
  --max-files 1 \
  --output-dir ./data/gdelt-mentions \
  --preview-lines 2 \
  --validate-structure \
  --expected-columns 16 \
  --quarantine-dir ./data/gdelt-mentions-quarantine \
  --log-level INFO \
  --log-file ./logs/gdelt-mentions-fetch.log \
  --pretty

Built-in Robustness

• Apply retry with exponential backoff on transient HTTP/network failures.
• Respect Retry-After when present on retriable responses.
• Throttle request frequency with a minimum interval between requests.
• Enforce --max-files safety cap (GDELT_MAX_FILES_PER_RUN) to prevent accidental bulk pulls.
• Validate datetime format and range boundaries before remote calls.
• Validate transport and structure after download:

- ZIP CRC/integrity check - UTF-8 strict decoding check - Tab column-count check (default 16) - Optional bad-line issue quarantine (--quarantine-dir)

• Emit JSON results while writing operational logs to stderr and optional log file.

Scope Decision

• Keep one concrete file-table fetch implementation: Global Mentions (*.mentions.CSV.zip).
• Keep atomic operations only; do not add internal scheduler/polling loops.

References

• references/gdelt-data-sources.md
• references/gdelt-limitations.md
• references/gdelt-schema.md
• references/env.md
• references/openclaw-chaining-templates.md

Script

• scripts/gdelt_mentions_fetch.py

OpenClaw Invocation Compatibility

• Keep skill trigger metadata in name, description, and agents/openai.yaml.
• Invoke in prompts with $gdelt-mentions-fetch.
• Keep the skill atomic: only resolve/fetch on demand.
• Use script parameters for fetch conditions (--mode range --start-datetime --end-datetime).
• If you need polling, let OpenClaw orchestrate repeated invocations externally, not inside this skill.

OpenClaw Prompt Templates

Use these templates directly in OpenClaw and only replace bracketed placeholders.

1. Recon (latest availability)

Use $gdelt-mentions-fetch.
Run:
python3 scripts/gdelt_mentions_fetch.py resolve-latest --pretty
Return only the JSON result.

1. Fetch (historical window, dry-run first)

Use $gdelt-mentions-fetch.
Run:
python3 scripts/gdelt_mentions_fetch.py fetch \
  --mode range \
  --start-datetime [YYYYMMDDHHMMSS] \
  --end-datetime [YYYYMMDDHHMMSS] \
  --max-files [N] \
  --dry-run \
  --pretty

Then run without --dry-run using:
  --output-dir [OUTPUT_DIR]
  --validate-structure
  --expected-columns 16
  --quarantine-dir [QUARANTINE_DIR]
Return only the JSON result.

1. Validate (download quality gate)

Use $gdelt-mentions-fetch.
Run:
python3 scripts/gdelt_mentions_fetch.py fetch \
  --mode latest \
  --max-files 1 \
  --output-dir [OUTPUT_DIR] \
  --validate-structure \
  --expected-columns 16 \
  --quarantine-dir [QUARANTINE_DIR] \
  --pretty
Check validation.issue_count, decode_error_count, column_mismatch_count.
Return JSON plus one-line pass/fail verdict.