Skip to content

skillferry 0.1.0

Latest

Choose a tag to compare

@GreenLv GreenLv released this 20 Aug 14:52
· 14 commits to main since this release

0.1.0 — 2026-08-20

First release: the portable agent-workspace core (roadmap stage A).

Added

  • workspace.toml schema v1: skills, instructions (marker/copy/include),
    MCP registry (secret-reference-only env), extension manifest, orthogonal
    overlays (base < target < platform < host < local), [protect]
    declarations, with schema-level rejection of literal secrets, path
    traversal, symlinks, and protect mis-declarations.
  • Adapters for Codex, Claude Code, and DeepSeek Harness with
    evidence-backed portability grades
    (native / translated / degraded / manual / unsupported).
  • CLI: init, import --from codex|claude
    (PORTABLE / LOCAL-ONLY / SENSITIVE / UNKNOWN classification),
    plan, apply (with --resolve id=adopt|overwrite|keep-local),
    doctor (exit codes 0/1/2/3), status, export --shareable,
    migrate --from codex-profile-sync.
  • Hash-based ownership ledger: create/update/adopt/delete semantics,
    source-change vs local-edit distinction, conflict blocking, per-target
    rollback, raw + redacted backups.
  • Seed skills setup-skillferry and release-checklist, and the runnable
    examples/starter-workspace (parity-checked in CI).
  • Public-tree auditor (scripts/audit_public_tree.py) and workspace
    validator (scripts/validate_workspace.py).
  • Test suite (68 tests) and CI matrix: 3 OS × Python 3.11–3.13 plus a
    release workflow.
  • Documentation: bilingual README, THREAT_MODEL, AGENT_MATRIX,
    PORTABILITY_CONTRACT, COMPARISON, MIGRATION, and the macOS native
    acceptance record.

Docs

  • Added a documentation index for the docs/ tree
    (docs/README.md) and Chinese translations of the core
    documents (docs/zh-CN/, CONTRIBUTING.zh-CN.md,
    SECURITY.zh-CN.md).
  • READMEs gained tables of contents, a grade glossary, and a documentation
    map; the Chinese README was resynced with the English one.

Fixed

  • Hardened multiline secret scanning, imported-skill inspection, opaque-binary
    rejection, and legacy migration against symlink traversal.
  • Made rollback restore partial writes in the failing target group and preserve
    deleted files, with redacted views generated from the actual backup bytes.
  • Persisted ledger-only adopt / keep-local resolutions.
  • Reconciled whole-skill and MCP-server removals across Codex, Claude Code, and
    DeepSeek Harness without deleting locally modified owned content silently.
  • Expanded the regression suite from 68 to 79 tests.
  • Pinned the build backend to hatchling<1.32 so distributions keep
    Metadata-Version: 2.4, which passes twine check --strict (hatchling
    1.32+ emits 2.5, rejected by twine 6.2 / packaging 26.x).