feat(skills): add prompt review skill (PromptSentinel v1.2)

[alexeyv]

Summary

• Adds bmad-os-review-prompt skill that reviews LLM workflow step prompts for known failure modes
• PromptSentinel v1.2: 17-item failure mode catalog (silent ignoring, negation fragility, scope creep, etc.)
• Three parallel review tracks: adversarial sub-agent, catalog scan + execution simulation, and path tracing
• Structured output format with severity scoring and concrete mitigations

Test plan

• Run skill against a sample BMAD workflow prompt and verify output format
• Confirm sub-agent spawning works for Track A and Track C
• Validate all 17 failure modes are checked in Track B
• Test INPUT_NOT_A_PROMPT guard with non-prompt input

[github-actions]

@coderabbitai review

[augmentcode]

🤖 Augment PR Summary

Summary: Adds a new Claude skill (bmad-os-review-prompt) to audit LLM workflow-step prompts using the PromptSentinel v1.2 failure-mode framework.

Changes:

• Introduces .claude/skills/bmad-os-review-prompt/SKILL.md with a full, self-contained prompt-auditing procedure.
• Defines a 4-step workflow: input validation, dependency/context inventory, three parallel review tracks, then merge + synthesis.
• Implements three analysis tracks: adversarial sub-agent review, catalog scan + execution simulation, and a path-tracing sub-agent.
• Adds a 17-item failure-mode catalog covering common prompt reliability risks (ignoring, negation fragility, scope creep, missing success criteria, etc.).
• Provides a strict Markdown output template with severity counts, quoted locations, risk framing, and rewritten mitigations.

Technical Notes: The skill is aimed at production-scale prompt execution across heterogeneous frontier models and emphasizes deterministic, mergeable findings with concrete rewrites.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 5 suggestions posted.

Comment augment review to trigger a new review at any time.