Commit graph

6 commits

Author SHA1 Message Date
avaclaw1
2aedbd3d2f feat(memory): advisory hygiene checks for durable content quality
The memory tool schema already tells the model not to save task progress,
session outcomes, or completed-work logs. In practice models still over-save
under completion bias -- writing 'deployed X on DATE' or 'rotated Y on DATE'
entries that have no durable value and crowd out genuinely useful facts.

This adds a lightweight defense-in-depth backstop: advisory hygiene checks
that run on every 'add' and surface a 'hygiene_warning' field on the
response. Three checks:

1. Session-log phrasing: regex patterns for dated completion verbs
   (deployed/rotated/retired/installed/completed/shipped/resolved/fixed
   followed by YYYY-MM-DD), 'views (DATE)' snapshots, and 'as of DATE'
   patterns. These are strong tells for ephemeral content.

2. Size pressure: warns once projected usage crosses 75% of the limit,
   so pruning advice surfaces BEFORE the hard block at 100%.

3. Near-duplicate detection: fuzzy-matches new content against existing
   entries using difflib SequenceMatcher. Similarity >= 0.75 suggests
   the caller should use 'replace' instead of 'add'.

All checks are advisory (non-blocking) -- the entry is still persisted.
Existing injection/exfil hard-blocks (_scan_memory_content) are unchanged
and take priority. The schema's existing DO-NOT-SAVE guidance is also
unchanged; this is belt-and-suspenders runtime reinforcement.

Tests: 16 new (10 direct unit tests for _check_hygiene, 6 integration
tests via MemoryStore.add). All 49 memory tests pass.
2026-04-24 19:36:40 +08:00
Teknium
8d023e43ed
refactor: remove dead code — 1,784 lines across 77 files (#9180)
Deep scan with vulture, pyflakes, and manual cross-referencing identified:
- 41 dead functions/methods (zero callers in production)
- 7 production-dead functions (only test callers, tests deleted)
- 5 dead constants/variables
- ~35 unused imports across agent/, hermes_cli/, tools/, gateway/

Categories of dead code removed:
- Refactoring leftovers: _set_default_model, _setup_copilot_reasoning_selection,
  rebuild_lookups, clear_session_context, get_logs_dir, clear_session
- Unused API surface: search_models_dev, get_pricing, skills_categories,
  get_read_files_summary, clear_read_tracker, menu_labels, get_spinner_list
- Dead compatibility wrappers: schedule_cronjob, list_cronjobs, remove_cronjob
- Stale debug helpers: get_debug_session_info copies in 4 tool files
  (centralized version in debug_helpers.py already exists)
- Dead gateway methods: send_emote, send_notice (matrix), send_reaction
  (bluebubbles), _normalize_inbound_text (feishu), fetch_room_history
  (matrix), _start_typing_indicator (signal), parse_feishu_post_content
- Dead constants: NOUS_API_BASE_URL, SKILLS_TOOL_DESCRIPTION,
  FILE_TOOLS, VALID_ASPECT_RATIOS, MEMORY_DIR
- Unused UI code: _interactive_provider_selection,
  _interactive_model_selection (superseded by prompt_toolkit picker)

Test suite verified: 609 tests covering affected files all pass.
Tests for removed functions deleted. Tests using removed utilities
(clear_read_tracker, MEMORY_DIR) updated to use internal APIs directly.
2026-04-13 16:32:04 -07:00
Teknium
8a384628a5
fix(memory): profile-scoped memory isolation and clone support (#4845)
Three fixes for memory+profile isolation bugs:

1. memory_tool.py: Replace module-level MEMORY_DIR constant with
   get_memory_dir() function that calls get_hermes_home() dynamically.
   The old constant was cached at import time and could go stale if
   HERMES_HOME changed after import. Internal MemoryStore methods now
   call get_memory_dir() directly. MEMORY_DIR kept as backward-compat
   alias.

2. profiles.py: profile create --clone now copies MEMORY.md and USER.md
   from the source profile. These curated memory files are part of the
   agent's identity (same as SOUL.md) and should carry over on clone.

3. holographic plugin: initialize() now expands $HERMES_HOME and
   ${HERMES_HOME} in the db_path config value, so users can write
   'db_path: $HERMES_HOME/memory_store.db' and it resolves to the
   active profile directory, not the default home.

Tests updated to mock get_memory_dir() alongside the legacy MEMORY_DIR.
2026-04-03 13:10:11 -07:00
teknium1
5319bb6ac4 fix: tighten memory and session recall guidance
Remove diary-style memory framing from the system prompt and memory tool
schema, explicitly steer task/session logs to session_search, and clarify
that session_search is for cross-session recall after checking the current
conversation first. Add regression tests for the updated guidance text.
2026-03-14 11:36:47 -07:00
teknium1
8253b54be9 test: strengthen assertions in skill_manager + memory_tool (batch 3)
test_skill_manager_tool.py (20 weak → 0):
  - Validation error messages verified against exact strings
  - Name validation: checks specific invalid name echoed in error
  - Frontmatter validation: exact error text for missing fields,
    unclosed markers, empty content, invalid YAML
  - File path validation: traversal, disallowed dirs, root-level

test_memory_tool.py (13 weak → 0):
  - Security scan tests verify both 'Blocked' prefix AND specific
    threat pattern ID (prompt_injection, exfil_curl, etc.)
  - Invisible unicode tests verify exact codepoint strings
  - Snapshot test verifies type, header, content, and isolation
2026-03-05 18:51:43 -08:00
0xbyt4
0ac3af8776 test: add unit tests for 8 untested modules
Add comprehensive test coverage for:
- cron/jobs.py: schedule parsing, job CRUD, due-job detection (34 tests)
- tools/memory_tool.py: security scanning, MemoryStore ops, dispatcher (32 tests)
- toolsets.py: resolution, validation, composition, cycle detection (19 tests)
- tools/file_operations.py: write deny list, result dataclasses, helpers (37 tests)
- agent/prompt_builder.py: context scanning, truncation, skills index (24 tests)
- agent/model_metadata.py: token estimation, context lengths (16 tests)
- hermes_state.py: SessionDB SQLite CRUD, FTS5 search, export, prune (28 tests)

Total: 210 new tests, all passing (380 total suite).
2026-02-26 13:27:58 +03:00