hermes-admin/hermes-agent
1
0
Fork 0
Code Issues Pull Requests Actions 31 Packages Projects Releases Wiki Activity
39,876 Commits 1 Branch 45 Tags
v2026.9.21
Commit Graph

6 Commits

Author SHA1 Message Date
teknium1
330004e597 fix(agent): keep trajectory serialization inside the best-effort handler 
A non-JSON value used to log 'Failed to save trajectory' and return; moving
json.dumps above the try made it raise to the caller.
 2026-09-12 08:25:14 -07:00
shafdev
bd70d7387a fix(agent): lock save_trajectory() appends so concurrent processes cannot corrupt the JSONL 
Gateway sessions and batch workers append to the same default
trajectory_samples.jsonl / failed_trajectories.jsonl with a plain open("a") +
write(); concurrent writers interleaved mid-object and the file stopped
parsing (#12684). The append now holds an exclusive lock for write+flush:
flock on POSIX, a 1-byte msvcrt.locking range on Windows.

Tests: a foreign process holding the lock must block the append (red on
base); six processes appending oversized entries all land parseable.

Fixes #12684. Salvaged from #12685 by @shafdev; Windows arm and test trim ours.
 2026-09-12 08:25:14 -07:00
Teknium
63f1950b4f refactor(agent/E_session): by-hand docstring/comment compaction across 9 files; transcript_repair row lookup helper; activity snapshot literal 2026-09-02 22:23:45 -07:00
Teknium
2a71fd4e03 refactor(agent/replay_cleanup,trace_upload,transcript_repair,session_activity,trajectory): collapse duplicate branches, compact docstrings 2026-09-02 18:31:56 -07:00
Teknium
c408601937 refactor(agent/review): simplify curator, background_review, verify, insights, title and learning modules (-22% LOC) 
Cluster: agent/{curator,curator_backup,background_review,review_engine,
review_idle_queue,insights,learning_graph,learning_graph_render,
learning_mutations,learn_prompt,verification_evidence,verification_stop,
verify_hooks,side_question,title_generator,turn_summary,
manual_compression_feedback,trajectory,moa_trace,trace_upload,verify/*}.
13662 -> 10693 LOC (-2969, -21.7%), behavior-neutral.

- Dead code: 27 private helpers with zero references removed
  (_auto_title_session, _resolve_review_model, _parse_make_targets,
  _filter_verifiable_paths, _find_subsequence, _is_under_root/_temp_dir,
  _merge_runs, learning_graph_render bucket/period/node helpers,
  _memories_dir/_memory_local_index/_node_detail, _cron_jobs_file,
  _retention_cutoff, _scope_for_args, _clean_token, _count_diff_lines,
  _ordered_verbs, _hermes_meta, _iter_skill_files).
- Unified helpers: _read_config_section (curator + curator_backup),
  _write_file/_write_json (4 curator report writers), _msg_text
  (background_review <- side_question), _report_failure/_notify_title
  (title_generator instant/auto paths), _is_under (verification_evidence),
  _scoped SQL pair builder + _query (insights), _optional_lock
  (background_review), verify.recipes table-driven detection.
- if/elif routing -> dict dispatch: side_question role labels,
  curator_backup summary bits, learning_graph_render buckets, insights
  section rendering, verify recipe pickers.
- Redundant defensive layers, single-use wrappers and verbose narrative
  comments collapsed; every non-obvious WHY/invariant kept in compact form.

Verification: parity.py (all REMOVED symbols zero-ref), import smoke for
every module + cli/run_agent/gateway.run/hermes_cli.main/
agent.conversation_loop/tui_gateway.server, old-vs-new fuzz parity on all
shared pure functions, SQL trace parity for insights and
verification_evidence, cluster tests 1354 passed / 0 failed (46 files).
 2026-09-02 13:30:25 -07:00
teknium1
9123cfb5dd Refactor Terminal and AIAgent cleanup 2026-02-21 22:31:43 -08:00