Get Job Memory

#2303 (R-038, Frederick 2026-08-18): "Rooster shouldn't decide what to remember. He should just know what bank to read for the job he's doing, because Rooster does many jobs." This route IS that wire: pass the calling job's own skill_key (the SAME constant departments. department_for already resolves -- e.g. domain.CS_REPLY_SKILL_KEY, departments.TECHNICAL_TRIAGE_SKILL_KEY) and it reads every knowledge bank THAT job's own department declares (departments.DEPARTMENT_PROFILES[dept].knowledge_graphs) -- never a bank the caller picks by hand, and never fewer than the department declares. A department that declares more than one bank (Technical: company decisions, software root causes, AND equipment fixes -- three graphs, since Zep searches one per call) is read across all of them and merged, each fact tagged with which bank it came from (job_memory.read_job_banks).

Gated on the same rooster:today:direction grant get_organization_ memory above already reuses for exactly the same reason: this is a cross-surface Rooster-agent read, not a human UI page.

Raises 404 for a skill_key this build does not recognize (departments.UnknownSkill) -- fails closed rather than guessing the widest department. org_id is honored only for a department whose own profile reaches org graphs (today: CS-Gov); every other department ignores it, exactly as departments.graphs_for_department documents. configured=False in the response means Zep itself is unconfigured (ZEP_API_KEY unset) -- an environment fact, not an error.