Source Etl Query

source-etl-query

Use this skill to query the host-side REST bridge served by source-etls. This is the mirror path, not the live GitHub REST path.

When to use

• Inspect mirrored GitHub issues or pull requests when the user explicitly asks for mirror data, historical mirror state, or ETL freshness checks
• Inspect mirrored GitHub discussions
• Inspect mirrored NVIDIA forum topics
• Build digests or comparisons that should use the hourly ETL mirror

Use github-readonly-live instead when the user asks for current live GitHub issues, PRs, commits, branches, README, or repository contents from a repository selected from $GITHUB_READONLY_REPOS or the legacy $GITHUB_READONLY_REPO fallback.

Access model

• Use the helper script in this skill.
• Prefer the helper scripts over custom REST queries
• The sandbox should treat this bridge as the default source for GitHub discussions and NVIDIA forum data in the NVIDIA setup path.
• The mirror and the live GitHub allowlist may cover different repositories. Keep mirror facts and live GitHub facts labeled separately.

Required Environment

• SOURCE_ETL_API_URL, or
• SOURCE_ETL_API_HOST plus SOURCE_ETL_API_PORT

Procedure

Always run these commands via the terminal tool — do not invoke source-etl-query as a named skill tool.

1. Query the mirrored source through the REST bridge

/usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py github-discussions --limit 20
/usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py forum-topics --limit 20
# Historical mirror checks only; use github-readonly-live for current issues/PRs.
/usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py github-issues --limit 20
/usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py github-prs --limit 20

2. Narrow the result set when needed

/usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py github-discussions --search <keyword> --limit 10
/usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py forum-topics --search <keyword> --limit 10

3. Interpret the results and handle empty or mismatched data

The ETL is configured to mirror a specific GitHub repo and forum tag — these are set by whoever deployed this sandbox, and may differ from what the user is asking about.

If the query returns zero rows:

• The database may be empty because the ETL has not completed its first sync yet (it runs on an interval, typically hourly). Tell the user: "The mirror appears empty — the ETL may not have completed its first sync yet. Try again in a few minutes."
• Alternatively the ETL target repo or forum tag may not match what the user is asking about (see below).

If results exist but are not relevant to the user's question:

• Run a broad unfiltered query first to show what IS in the database:

  /usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py github-discussions --limit 5
  /usr/bin/python3 /sandbox/.hermes-data/skills/source-etl-query/scripts/query_source_etl.py forum-topics --limit 5
  

• Then tell the user what repo/topics the mirror actually contains, e.g.: "The ETL mirror contains discussions from NVIDIA/NemoClaw and forum topics tagged nemoclaw. If you need mirror data from a different repo or topic, the ETL target would need to be reconfigured."

Do not fall back to live NVIDIA forum requests — the sandbox has no egress to forum HTML. For GitHub, switch to github-readonly-live only when the user needs current live data from an allowed repository; otherwise explain the ETL mirror scope or freshness limitation.

Pitfalls

• The mirror is refreshed hourly, so it is not guaranteed to match the live source instantly.
• The ETL scope is configured per-deployment — forum topics follow the configured tag, and GitHub mirror rows follow the configured repo. Neither covers the whole of GitHub or the entire forums site.
• Do not describe mirror data as live. If combining it with live GitHub REST data, say which facts came from the mirror and which came from live REST.