Skip to content

feat(plugins): fail-closed required enforcers + scheduled-run/model-spend governance - #1

Merged
ElystrumCore merged 18 commits into
mainfrom
agent/required-run-enforcers
Jul 19, 2026
Merged

feat(plugins): fail-closed required enforcers + scheduled-run/model-spend governance#1
ElystrumCore merged 18 commits into
mainfrom
agent/required-run-enforcers

Conversation

@ElystrumCore

Copy link
Copy Markdown
Owner

Summary

Companion to agent-lineage PR NousResearch#14/NousResearch#15: generic required-plugin support (plugins.required, register_required_hook, --require-plugin) with fail-closed startup, plus the pre_api_request/post_api_request/api_request_error and pre_run_start/post_run_end hook seams so cron/kanban execution and model calls are admitted/metered by the agent-lineage policy worker before spend. A missing, disabled, crashed, timed-out, or malformed mandatory enforcer aborts startup or blocks before execution.

Verification

  • Affected-runtime suite on the branch: 652 passed (2026-07-19).
  • Live-proven against the agent-lineage HermesPolicyBridge/DurableRunController seam (two-phase reservations, cron/kanban admission, durable run binding via AGENT_LINEAGE_RUN_ID).

🤖 Generated with Claude Code

CN-CLI added 17 commits July 12, 2026 18:38
…s repeating a prior action; single-action starter goal 1

Empirical fine-tune on qwen3:8b @ Ollama :11436. Progression: v1 (7B) hallucinated a
function-name action; v2 copied the prompt's e.g. example verbatim; v3 (this) tailors a
varied natural-language action to the goal. Remaining blocker is the ACTUATOR (Hermes
agent execution), not the brain: qwen3:8b's thinking mode pollutes Hermes tool-calls
(/think leaked into 'ls /think') and an 8B mis-drives the tool protocol. Config is sound
(terminal=local, cwd=workspace); the fix is a competent acting model, not a config knob.
@cursor

cursor Bot commented Jul 19, 2026

Copy link
Copy Markdown

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

Both-sides resolutions: cron run_one_job keeps upstream execution
tracking (create/finish_execution, finished on the policy-deny early
return too) alongside the governed session id + post_run_end observers;
kanban _default_spawn keeps the HERMES_TUI strip and the pre-Popen
required-policy admission; _prepare_agent_startup keeps the --yolo
chokepoint ahead of --require-plugin env inheritance; both sides' tests
retained (deny test now patches create/finish_execution).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@ElystrumCore
ElystrumCore merged commit d61f75e into main Jul 19, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant