Skip to content

Releases: HaliosAI/halios

CLI + Agent Skill 2.0.9

Choose a tag to compare

@sandeepbele sandeepbele released this 29 Aug 17:55
6478e60

CLI + Agent Skill 2.0.9 keeps terminal quota failures actionable without letting coding agents retry or broaden the task automatically.

Changed

  • Format structured HTTP 402 usage-limit responses in the CLI with the affected meter, usage, server remediation, and billing URL.
  • Stop evaluation retry loops when a run is marked incomplete_quota.
  • Derive evaluation-run billing links from the configured Halios UI origin.
  • Present BYOK only for managed-AI-token exhaustion; check-result and trace-ingest limits require pay-as-you-go or the monthly reset.
  • Restore the bounded Skill workflow: setup stops after one verified k=1 smoke scenario, full runs preserve the user's explicit trial count, and run requests do not authorize behavioral fixes or commits.
  • Align Skill, CLI, README, connect workflow, and bundled CI-gate minimum versions at 2.0.9.

Verification

  • Public PR #12 passed Skill installation and distribution checks.
  • CLI tests passed on Python 3.10, 3.11, 3.12, 3.13, and 3.14.
  • Release artifacts are installed and verified on Linux, macOS, and Windows before PyPI publication.

CLI + Agent Skill 2.0.8

Choose a tag to compare

@sandeepbele sandeepbele released this 29 Aug 04:20
3952552

CLI and Agent Skill 2.0.8

This release improves evidence-aware evaluation authoring while keeping Halios setup bounded and transparent.

Highlights

  • Records unresolved evidence and coverage gaps in optional local .halios/discovery.yml notes.
  • Reports discovery gaps separately from executable-suite validity and evaluation outcomes.
  • Adds concise, matched RAG check and scenario examples for retrieval evidence, faithfulness, relevance, correctness, citations, and honest evidence limits.
  • Distinguishes fixed-question coverage from repeated trials: -k repeats stable scenarios; it does not generate new questions.
  • Supports plain-language “expand the scenarios” requests, adapting to fixed-question datasets or conversational situations.
  • Keeps setup bounded to one smoke scenario unless a broader run is explicitly requested.

Package boundary

This is the coupled haliosai-cli and Halios Agent Skill release. It does not release the separate haliosai Python SDK.

CLI + Agent Skill 2.0.7

Choose a tag to compare

@sandeepbele sandeepbele released this 27 Aug 19:16
adffae1

CLI + Agent Skill 2.0.7

  • Make first-time CLI setup a bounded, platform-adaptive workflow using isolated user-level tooling.
  • Explain installation to the user before proceeding, including reuse across projects and separation from application dependencies.
  • Bundle a complete eval.example.yml with deterministic structure/value checks, message and tool scoping, and focused semantic judges.
  • Clarify that regex and string matching should not stand in for semantic correctness; use an LLM judge when meaning, context, or valid paraphrasing matters.
  • Add automated validation of the bundled example.

Implemented in #6; coupled version alignment in #7.

Install

npx skills add HaliosAI/halios --skill halios

For direct CLI use:

uv tool install 'haliosai-cli==2.0.7'

This releases haliosai-cli and the Halios Agent Skill. The separate haliosai Python SDK is unchanged.