PR #23208: feat(kanban): stamp originating ACP session_id on tasks

• Repository: NousResearch/hermes-agent
• Author: awizemann
• State: open | merged: False
• Link: https://github.com/NousResearch/hermes-agent/pull/23208

Description (problem / solution / changelog)

Closes #23199.

Summary

Track which chat session created a kanban task so clients can render a per-session board without falling back to tenant + time-window heuristics.

What changed

• Schema — tasks gains a nullable session_id TEXT column with an index. Migration follows the existing _migrate_add_optional_columns pattern (the max_retries precedent), so legacy DBs upgrade in place on the next init_db call.
• ACP propagation — acp_adapter/server.py exposes the originating session id to the agent loop via HERMES_SESSION_ID, mirroring the existing save/set/restore lifecycle around HERMES_INTERACTIVE. Restoring is important so a re-used executor thread doesn't leak one session's id into the next session's tools.
• Tool — tools/kanban_tools.py _handle_create reads HERMES_SESSION_ID (with an explicit session_id arg override) and passes it to kb.create_task, parallel to the existing HERMES_TENANT fallback. NULL on CLI/dashboard paths so behaviour is unchanged for non-ACP callers.
• CLI — hermes kanban list accepts --session <id>. JSON output now includes session_id. The flag ANDs with existing --tenant / --assignee filters.

Testing

• 8 new tests across the kanban DB, tools, CLI, and ACP server suites:
  • kanban_db stamping + filtering, including a tenant×session compose case.
  • kanban_create env propagation, explicit-arg override, and absent-env defaults-to-NULL.
  • CLI --session filter + JSON output exposes session_id.
  • ACP _run_agent sets HERMES_SESSION_ID during the call and restores the prior value (covers the nested-loop case so an outer HERMES_SESSION_ID survives an inner prompt).
• Existing kanban + ACP suites unchanged: 402 passing locally (tests/hermes_cli/test_kanban_*.py, tests/tools/test_kanban_tools.py, tests/acp/, tests/acp_adapter/).
• The pre-existing concurrency-migration test (test_migrate_add_optional_columns_tolerates_concurrent_migration) had its already-migrated schema fixture extended to include the new column, so the "migration is no-op when columns exist" guarantee still covers session_id.

Backwards compatibility

• Legacy DBs migrate in place; existing rows get session_id = NULL, which is the correct default.
• Non-ACP creation paths (CLI, dashboard, legacy hosts that don't set the env) continue to produce session_id = NULL rows. The CLI and JSON shape changes are additive — no flag is removed, no JSON key is renamed.
• The HERMES_SESSION_ID env var is only read on the kanban-create code path; if a client wants to opt their tools out, simply not setting the env produces today's behaviour.

Use case (downstream)

We're building Scarf, a macOS GUI for Hermes. While a user has a /goal running, we want to show a Kanban board scoped to "tasks this chat is producing." Without session linkage we have to approximate via tenant + time-window — which over-collects for concurrent chats on the same project and under-collects for chats with no tenant. With this PR landed, Scarf can pass --session <acp-session-id> for a precise filter.

🤖 Generated with Claude Code

Changed files

• acp_adapter/server.py (modified, +14/-1)
• hermes_cli/kanban.py (modified, +5/-0)
• hermes_cli/kanban_db.py (modified, +40/-3)
• tests/acp/test_server.py (modified, +74/-0)
• tests/hermes_cli/test_kanban_cli.py (modified, +42/-0)
• tests/hermes_cli/test_kanban_db.py (modified, +72/-1)
• tests/tools/test_kanban_tools.py (modified, +69/-0)
• tools/kanban_tools.py (modified, +5/-0)