ADR-0024 — Session journal and transcript search instead of a database

Status: Accepted, partly superseded by ADR-0039 (a database for the pages read online)

Context

Conversations with the Product Owner are long and get compacted; decisions end up in the repository, but the path to them (what was asked, discussed, refused) is scattered. The Product Owner suggested a local database where the agent stores everything it may need. Long-running agents keep coherent through structured notes outside the context window, reloaded just in time (REF-ANTHROPIC-CONTEXT). Claude Code already stores every conversation locally as a JSONL transcript.

Decision

1. A session journal, docs/process/journal.md: per session, the requests (paraphrased), decisions and where they were recorded, with #tags for grep. Written by the agent at the end of a session or before a pause.
2. The transcripts are the full archive: searched with grep / jq when the journal and the repository are not enough (docs/tooling/agent.md).
3. No database (SQLite, MCP memory server): it would duplicate the repository, not be reviewed in pull requests, and need its own tooling; Markdown and grep are what the agent reads fastest.

Consequences

• One file to skim after a compaction, besides the session-start hook.
• The journal is public like the rest of the repository: no personal data, only project matters.

References

REF-ANTHROPIC-CONTEXT, ADR-0021, ADR-0023