pipeline-state

Durable pipeline state is maintained by persisting per-run variables and per-phase outputs in a JSON file under `.ai_log/`, enabling downstream agents to recover required data by lookup instead of receiving inlined dumps.

The best way to pass state is using a file-backed handoff bus that preserves per-run variables and evidence paths, allowing agents to retrieve exactly what they need by lookup to survive context compression.

Initialize pipeline state using `pipeline-state.sh init <session_id>`, then add data with `set <session_id> <field> <value>` or append to lists with `append <session_id> <field> <value>`.

Yes, you need jq installed for JSON manipulation, as the pipeline state management system requires it to parse and update the file-backed state alongside Bash for script execution.

Pipeline state management coexists with human-readable reports by offloading structured data to `.ai_log/` without duplicating content, ensuring both agents and humans access the formats they need.

Multi-agent handoff fails because inlined state dumps are lost during context compression, a problem solved by persisting per-phase outputs and variables in a durable JSON file for reliable lookup.