Skip to content

v1.26.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 05 May 08:35
· 301 commits to main since this release
v1.26.0
d7a642b
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Highlights

This release makes gentle-ai's emitted personas instruct the agent to consult and invoke skills proactively — without the user authoring any trigger table by hand. Closes #435.

Before: skills were installed under ~/.claude/skills/ and friends, listed in <available_skills>, but the model rarely invoked them unless the user had hand-written a MANDATORY directive in their personal ~/.claude/CLAUDE.md.

After: every emitted persona ships a generic mandatory directive that delegates skill invocation to the agent's native discovery mechanism.

What changed

  • Persona directive injected into 6 assets — claude/persona-gentleman.md, opencode/persona-gentleman.md, generic/persona-gentleman.md, generic/persona-neutral.md, kiro/persona-gentleman.md, kimi/persona-gentleman.md. Claude variant references the built-in Skill tool; non-Claude variants reference reading SKILL.md natively.
  • Hardcoded skills auto-load table removed from those 6 personas (the old 2-row table that listed only go-testing + skill-creator and competed with <available_skills>).
  • Frontmatter hygiene — chained-pr/SKILL.md name: now matches its directory; skill-creator/SKILL.md no longer carries the non-standard allowed-tools: field.
  • Compatibility insurance — every description: > block scalar in the 21 emitted SKILL.md is now a single-line plain scalar, immune to parser fragility per anthropics/claude-code#9716.
  • Regression guards — new TestSkillFrontmatterIsLintClean (frontmatter rules) and TestPersonasContainContextualSkillLoadingDirective (asserts the directive shape across all 6 personas).

What this means for you

Re-sync your agents to pick up the new persona:

brew upgrade gentle-ai
gentle-ai sync

Your installed skills are discovered the same way — but agents will now consult and invoke them proactively. No manual ~/.claude/CLAUDE.md trigger table required.

Deferred (tracked, not in this release)

  • Behavioral verification of the directive — live-agent transcript or Claude API automation showing the model actually invokes skills proactively under the new persona. This release ships structural-only evidence.
  • Linter hardening for required license and metadata.author fields.

Full SDD trail

Available under openspec/changes/archive/2026-05-05-contextual-skill-loading/ — proposal, design, tasks, specs (A/B/C/D), apply-progress, verify-report, archive-report.

Assets 9
Loading