fix(session): reject kild open on active sessions

[Wirasm]

Summary

kild open on an already-active session unconditionally spawned a second agent process and overwrote agent_session_id with a new UUID. This created competing agents in the same worktree and permanently broke --resume for the original conversation.

Root Cause

open_session() had no guard checking session status before spawning — unlike create_session() which checks for AlreadyExists. The agents vec got a duplicate push and rotate_agent_session_id() silently discarded the live session ID.

Changes

File	Change
crates/kild-core/src/sessions/errors.rs	Add AlreadyActive error variant with actionable alternatives (attach, inject, stop+reopen)
crates/kild-core/src/sessions/open.rs	Add active-session guard before agent spawn with daemon liveness check
crates/kild-core/src/sessions/errors.rs	Add test for AlreadyActive error
crates/kild-core/src/sessions/open.rs	Add tests for guard logic (active+agents blocked, stopped allowed)

Design Decisions

• Daemon liveness check: For daemon sessions, sync_daemon_session_status() verifies the PTY is truly running. Stale-active sessions (daemon PTY died without kild stop) get synced to Stopped and can reopen normally.
• Terminal sessions trust stored status: No reliable PID-based liveness check for external terminal processes. Users can kild stop then kild open.
• --all unaffected: The open --all CLI path already filters for Stopped sessions only.

Testing

• cargo fmt --check passes
• cargo clippy --all -- -D warnings passes
• cargo test --all passes (all existing + 3 new tests)

Validation

cargo fmt --check && cargo clippy --all -- -D warnings && cargo test --all

Issue

Fixes #599

[Wirasm]

PR Review Summary

Reviewed by 5 specialized agents: code-reviewer, pr-test-analyzer, silent-failure-hunter, comment-analyzer, and self-reviewer.

---

Critical Issues (3 found)

#	Agent(s)	Issue	Location
1	code-reviewer, self-reviewer, silent-failure-hunter, comment-analyzer	Redundant save_session_to_file in stale-active path. sync_daemon_session_status() already persists the status change via patch_session_json_fields() (which preserves unknown fields from newer binary versions). The follow-up save_session_to_file() is a full JSON rewrite that can silently drop unknown fields — directly contradicting the safety guarantee established in list.rs:117-119. Additionally, the info! fires before the save attempt with no _failed counterpart on error. Fix: remove the redundant save entirely.	open.rs:125-132
2	self-reviewer	Guard blocks BareShell opens on active sessions. The guard fires for all OpenMode variants including BareShell. A bare shell doesn't spawn an agent and can't corrupt agent_session_id — the exact scenario the guard exists to prevent. Users running kild open my-branch --no-agent on an active session will get a false AlreadyActive rejection. Fix: gate guard on !matches!(mode, OpenMode::BareShell).	open.rs:95-98
3	comment-analyzer	open_session docstring contradicts new behavior. The docstring says "additive - doesn't close existing terminals" and "multiple agents can run in the same kild" — the guard now prevents exactly that. Fix: update docstring to reflect the new guard.	open.rs:35-41

Important Issues (3 found)

#	Agent(s)	Issue	Location
4	self-reviewer, pr-test-analyzer, code-reviewer	Tests validate boolean condition copy, not open_session(). Both guard tests manually evaluate session.status == Active && session.has_agents() without calling open_session(). If the guard is removed or inverted, these tests still pass. The stale-active daemon path (the subtlest logic) is entirely untested. Fix: add end-to-end test calling open_session() on an Active terminal session asserting Err(AlreadyActive).	open.rs:1136-1187
5	silent-failure-hunter	No log event for terminal liveness-skip branch. When a non-daemon session hits the guard, there's no log indicating the terminal path was taken (trusting stored status without liveness check). Makes debugging "why was my open blocked?" harder. Fix: add info! with core.session.open_terminal_liveness_skipped.	open.rs:108-110
6	code-reviewer	Test names don't follow <subject>_<expected_behavior> convention. Per CLAUDE.md.	open.rs, errors.rs

Suggestions (5 found)

#	Agent(s)	Suggestion	Location
7	pr-test-analyzer	Add kild stop assertion to test_already_active_error (one-liner)	errors.rs:230
8	comment-analyzer	Add comment explaining !sync_daemon_session_status() negation (returns true when changed to Stopped, so ! = still running)	open.rs:106
9	silent-failure-hunter	Log resume field in open_started event	open.rs:51-56
10	silent-failure-hunter	Log session_id in open_stale_active_synced event	open.rs:127-131
11	pr-test-analyzer, comment-analyzer	test_stopped_session_not_blocked claims "even if agents non-empty" but doesn't add agents — add one to prove the claim	open.rs:1176-1187

Strengths

• Error message is actionable — three concrete recovery paths (attach, inject, stop && open)
• Guard placement is correct — after worktree check, before any agent resolution or spawn
• Daemon liveness check reuses existing sync_daemon_session_status() infrastructure
• --all path correctly unaffected (already filters for Stopped)
• Structured logging follows {layer}.{domain}.{action}_{state} naming convention
• AlreadyActive error type is well-formed: error code, is_user_error(), actionable display

Verdict

NEEDS FIXES — 3 critical issues require changes before merge. The core approach is sound; the guard logic, error variant, and daemon liveness integration are correct. The critical fixes are:

1. Remove the redundant save_session_to_file (5-line deletion)
2. Gate the guard on !BareShell (1-line addition)
3. Update the docstring (text edit)

Recommended Action Order

1. Fix Critical Implement file-based persistence system #1 — remove redundant save
2. Fix Critical Add Ghostty terminal support (blocked by upstream -e parsing) #2 — gate on !BareShell
3. Fix Critical feat: implement cleanup tracking system #3 — update docstring
4. Fix Important feat: Hierarchical TOML Configuration System #4 — add end-to-end regression test
5. Fix Important Implement missing session management commands (destroy, cleanup, info) #5 — add terminal liveness-skip log
6. Apply suggestions as time permits