zishuq/caro/deepagents-thread-inspector

Deep Agents Thread Inspector

If LangSmith tooling is available for a traced thread, prefer it. Otherwise, use scripts/inspect_sessions.py instead of manually decoding database blobs. It opens the database read-only and deserializes the root message channel with LangGraph's strict MsgPack loader — reading the materialized messages from the latest checkpoint, or replaying writes in checkpoint order when that fast path is unavailable — and emits JSON.

Inspect local state

Resolve SKILL_DIR to the directory containing this SKILL.md; do not assume a user, project, or installation-specific location. Start with the smallest useful view:

python3 "$SKILL_DIR/scripts/inspect_sessions.py" THREAD_ID --mode latest-turn

A unique thread-ID prefix is accepted. Select another view when needed:

python3 "$SKILL_DIR/scripts/inspect_sessions.py" THREAD_ID --mode summary
python3 "$SKILL_DIR/scripts/inspect_sessions.py" THREAD_ID --mode transcript

Use --include-metadata only when run, repository, model, checkpoint, or LangGraph metadata matters. Use --max-content N to raise or lower the default 4,000-character limit per message, tool result, or tool-call argument.

If the user does not know the ID, list recent threads first:

python3 "$SKILL_DIR/scripts/inspect_sessions.py" --list 20

Pass --db PATH only for a non-default session store. The default is ~/.deepagents/.state/sessions.db; DEEPAGENTS_SESSIONS_DB can override it.

Explain the result

Synthesize the JSON rather than pasting it verbatim.

• State the user's request, the assistant's conclusion, and significant tool actions or failures.
• Distinguish stored facts from your interpretation.
• For the latest turn, describe only the final user message and subsequent activity unless earlier context is required to make it understandable.
• Mention truncation when a relevant record has content_truncated or args_truncated set.
• Surface reconstruction problems when the result includes a top-level warnings array (for example, a corrupt checkpoint, a skipped write, or malformed metadata) so conclusions are appropriately hedged.
• Do not expose unrelated credentials, tokens, personal data, or hidden reasoning that may appear in local records.

Safety

Keep inspection read-only. Do not deserialize an untrusted database: checkpoint deserialization is intended for trusted local Deep Agents state. Do not mutate or delete session rows unless the user separately and explicitly requests it.