feat(session): cleanup_period_days — opt-in session pruning

[emal-avala]

Summary

Adds [session] cleanup_period_days = N so long-running installs don't accumulate unbounded session files under ~/.config/agent-code/sessions/. When set, old session files are pruned at REPL startup. Absent / None preserves today's behavior (keep sessions forever).

Config

[session]
# Auto-delete session files whose `updated_at` is older than N days.
# Absent or None = keep forever.
cleanup_period_days = 30

Safety posture

• Default is no-op — no change for existing users; the feature is strictly opt-in.
• Conservative age check: files with missing or malformed updated_at are kept. We never delete a file whose age we can't determine.
• Non-fatal sweep: a failed remove_file is counted as kept and does not abort the rest of the pass (one locked file shouldn't stop the others).
• Strictly-less-than boundary: a file exactly at the threshold is kept, avoiding off-by-one surprises.
• Some(0) is a no-op: treated identically to None; reserved as a sentinel we can repurpose later without breaking existing configs.
• Non-JSON files skipped entirely — .tmp / .bak leftovers are never scanned or counted.

Surface in /doctor

• Enabled: ✓ Session cleanup: prune files older than 30 days
• Disabled: ! Session cleanup disabled — set [session] cleanup_period_days to prune old sessions

Tests (8)

Schema (3): default is None, TOML round-trip, a config without [session] keeps defaults.

Runtime (5): stale file removed, fresh kept, days = 0 is a no-op even with an ancient file, unparseable timestamp kept, non-JSON files skipped, exact threshold kept.

Full services::session, config::schema, and services::diagnostics suites pass. Clippy clean under -D warnings.

Test plan

• cargo test -p agent-code-lib --lib services::session
• cargo test -p agent-code-lib --lib config::schema
• cargo test -p agent-code-lib --lib services::diagnostics
• cargo build -p agent-code
• cargo clippy --workspace --tests --no-deps -- -D warnings

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.