Skip to content

feat(skill): decompose kane-cli skill + objectives cookbook + Monitor narration#61

Open
siddhant573 wants to merge 16 commits into
LambdaTest:mainfrom
siddhant573:docs/21-05-26
Open

feat(skill): decompose kane-cli skill + objectives cookbook + Monitor narration#61
siddhant573 wants to merge 16 commits into
LambdaTest:mainfrom
siddhant573:docs/21-05-26

Conversation

@siddhant573
Copy link
Copy Markdown
Collaborator

Summary

  • Decomposed the kane-cli skill from a monolithic 808-line SKILL.md into a thin 259-line entry point + six on-demand references/*.md files (testmd, parsing, debug, parallel, setup-and-config, objectives-cookbook). §1 (live narration + results table) is now first, not buried at line 353.
  • Added references/objectives-cookbook.md as a shared pattern catalog for both kane-cli run objectives and _test.md step bodies — covers all five checkpoint analyze methods (Visual / Textual-DOM / URL / Title / DevTools) and the five DevTools subdomains (Network, Console, Performance, Cookies, localStorage) the skill previously had no awareness of.
  • Added user-facing checkpoint docs at docs/user-guide/features/checkpoints/ (11 files) — source of truth for the cookbook.
  • Switched live narration to use Monitor in Claude Code (Bash is synchronous and can't stream stdout per event); Codex / Gemini fall back to Bash + post-run narration. Resolves the silent-during-runs bug.
  • Installer improvements: cli.js now stamps a VERSION file in each install target on every install, prints the version it installed, and version bumped to 1.1.0. Single source of truth = package.json.

Test plan

  • Fresh Claude Code session loads new ~/.claude/skills/kane-cli/SKILL.md (verify with cat ~/.claude/skills/kane-cli/VERSION1.1.0)
  • One-shot kane-cli run via skill: Starting browser task: … before launch, Monitor (not Bash) for the launch call, one narration line per progress event with no >2s silence, full 6-row results table after run_end
  • testmd authoring: skill Reads references/testmd.md AND references/objectives-cookbook.md (since step bodies are objectives)
  • Capability question ("what assertions can I do?"): skill Reads references/objectives-cookbook.md and surfaces checkpoint analyze methods + DevTools subdomains
  • Forced failure: §1.5 plain-language failure report (🔴 Failed / What happened / Likely cause / Suggested fix) and a Read of references/debug.md
  • Codex CLI: same skill content works with Bash + post-run narration fallback
  • Gemini CLI: imperative rules still apply (TodoWrite step is skipped gracefully)

🤖 Generated with Claude Code

siddhant573 and others added 16 commits May 26, 2026 13:10
…in testmd reference

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ssertions

Covers the five analyze methods (Visual, Textual/DOM, URL, Title, DevTools)
and the five DevTools subdomains (Network, Console, Performance, Cookies,
localStorage), with examples for Assert / Extract / If-Else checkpoint kinds
and the operator catalog.
…ds + refresh mapping

Mirrors the May 2026 SKILL.md restructure into the Kiro power:

- Adds an "Analyze methods — picking the right checkpoint" section to
  steering/kane-cli-run.md covering the five analyze methods (Visual,
  Textual/DOM, URL, Title, DevTools) and the five DevTools subdomains
  (Network, Console, Performance, Cookies, localStorage) with capture
  semantics and per-step reset rules. Closes the gap where Kiro had no
  awareness of any checkpoint type beyond generic "assert/verify".

- Rewrites the SKILL.md -> Kiro mapping table in
  integrations/docs/kiro-powers.md to reflect the new 6-section
  SKILL.md plus the six references/*.md files (objectives-cookbook,
  testmd, parsing, debug, parallel, setup-and-config).

- Notes explicitly that the Monitor/Bash launch decision in SKILL.md
  §1 is Claude-Code-specific; Kiro keeps its own narration model.
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