feat(cron): pass job name as session title for cron sessions

[waldmanz]

Problem

Cron job sessions previously had no title in the SQLite session store, causing the web UI and sessions list to display truncated system prompt text as session titles:

[SYSTEM: You are running as a scheduled cron job. DELIVERY: Your...
[SILENT]讨论
llm-wiki与AI效率

This happens because cron sessions bypass the gateway/CLI — they call agent.run_conversation() directly, so the maybe_auto_title() hook is never triggered. The web UI falls back to the first message content (session.preview) when title is NULL.

Solution

Pass the cron job's human-readable name through to SessionDB.create_session() so each cron session gets a meaningful title immediately: wiki-auto-ingest, Hermes Tips, AI docs vault updater, etc.

Changes

File	Change
hermes_state.py	Add optional title parameter to create_session()
run_agent.py	Accept session_title in AIAgent.init and forward it
cron/scheduler.py	Pass session_title=job_name when constructing the agent

Testing

• All 277 existing tests pass (cron, hermes_state, title_generator)
• Backward compatible — all new parameters default to None
• No migration needed — the title column already exists in the schema (v3 migration)

[waldmanz]

Thanks — I pushed a follow-up fix for the session-title behavior.

What changed

• Preserve cron session titles across repeated runs
  • Repeated runs of the same cron job now get unique lineage titles instead of colliding on the unique title index
  • Example:
    • wiki-auto-ingest
    • wiki-auto-ingest Support passing morph snapshot id #2
    • wiki-auto-ingest Architecture planning #3
• Sanitize titles on create_session()
  • This now uses the same title-cleaning rules as set_session_title()
  • Prevents raw whitespace/control-character titles from being inserted directly
• Preserve title in the late recovery path
  • If startup session creation is skipped/fails and run_agent later falls back to ensure_session(), the requested session title is still retained

Why this follow-up was needed
The initial version passed title=job_name into create_session(), but sessions.title is uniquely indexed.
That meant the first run of a recurring cron job worked, while later runs with the same job name could hit INSERT OR IGNORE, skip row creation, and then get recreated later without a title via ensure_session().

This follow-up fixes that by allocating a unique sanitized title before insert, and by threading the title through the ensure_session() recovery path too.

Tests added

• create_session with duplicate title gets numbered variant
• create_session sanitizes title input
• ensure_session preserves requested title
• run_agent flush path passes session_title into ensure_session

Validation

• Focused new regression tests pass
• Existing hermes_state title/ensure_session tests still pass
• Related run_agent tests still pass

Pushed in:

• a45a727 fix(cron): preserve unique session titles across repeated runs

[teknium1]

Thanks for addressing the cron-session naming gap. Current main already titles sessions in the scheduler finalizer, but the repeated-run edge case remains.

Problems

• cron/scheduler.py:3297 formats the suffix only through minutes. Two same-name runs completing in one minute generate the same title; hermes_state.py:2743-2745 raises ValueError, and cron/scheduler.py:3299-3300 logs and suppresses it, leaving the later session untitled.
• The PR targets an older persistence shape. Current session creation is centralized by hermes_state.py:1610-1668 and invoked by run_agent.py:598-607, so the creation-time plumbing needs adaptation rather than a direct cherry-pick.

Suggested changes

• Keep the current finalizer path and allocate a numbered fallback with SessionDB.get_next_title_in_lineage() (hermes_state.py:2852-2885) before set_session_title().
• Add a fixed-clock regression covering two same-name runs in one minute and assert both sessions retain distinct titles.

This is an automated hermes-sweeper review.

[teknium1]

Current main already has SessionDB.get_next_title_in_lineage() for numbered title allocation. Please rebase the duplicate-title behavior onto that existing API and the scheduler finalizer, rather than adding a second allocator in the session-creation path.

[teknium1]

Thanks for the contribution — this was the earliest PR (April 23) to target meaningful cron session titles, and the need was real.

Closing as superseded by main's design plus PR #66058: main now titles cron sessions in the scheduler's finally block after the run (deliberate — a create-time title would be overwritten by the agent's own INSERT bookkeeping, and loses the timestamp suffix that keeps repeated runs unique against the title index). #66058 hardened that path with collision dedup and a non-blank guarantee. The job-name-as-title outcome you wanted is what ships today; the mechanism just moved to the other end of the run.