Prompt JSONL Search for AI Coding Logs

Best-fit situations

• A coding agent plugin exports sessions as JSONL but the files are hard to inspect manually.
• A team needs to find every prompt that mentioned a customer, token, endpoint, or repository area.
• A prompt engineer wants to surface successful sessions by model and task type.
• A security owner wants redaction checks before logs are shared with contractors or clients.

Operating steps

1. Upload JSONL files or paste a representative export for indexing.
2. Parse message role, timestamp, model, token counts, tool calls, and output metadata.
3. Map each record to a session and assign repo, task, and outcome fields.
4. Run secret, domain, customer-data, and prompt-injection redaction checks.
5. Search and export the filtered result as a prompt pack or client appendix.

Common risks

• Ad hoc grep misses JSON structure, escaped content, multiline messages, and tool-call metadata.
• Large JSONL files can expose secrets when opened in shared docs or chat tools.
• Search results are less useful if sessions are not grouped by repo and task.
• Teams may preserve noisy logs without extracting reusable patterns.