feat: replace CLI arguments with interactive prompts in skills

[k-rister]

Summary

Replace command-line argument parsing with AskUserQuestion interactive prompts across crucible-tools skills. Users now get guided question-based workflows instead of needing to remember CLI argument syntax.

Changes

Skills updated (8 total):

1. activity-summary

   • Added Audience question (User/Developer mode with filtering)
   • Added Author scope question (Me/Specific/All)
   • Added Date range question (1w/2w/1m/Custom)
   • Enhanced key themes with sub-bullets for multi-subtopic themes
   • Added Jira ticket linking requirement

2. open-prs

   • Added Author filter question (All/Me/Specific)

3. dev-activity

   • Added Time range question (26/52/13 weeks or Custom)

4. workflow-status

   • Added Repository filter question (All/Specific repos)
   • Added Runner summary question (Yes/No)

5. new-repo

   • Added License question (apache-2.0/mit/gpl-3.0)
   • Added Repository details question (name|description format)

6. pr-review

   • Added PR selection question (Number/URL/Repo+Number)
   • Added Prior findings question (None/Exclude)

7. architecture-review

   • Added Scope question (All/7 individual dimensions)
   • Added Depth question (Survey/Deep)
   • Added Repo focus question (System-wide/Specific)

8. codebase-audit

   • Added Scope question (All/Core/Benchmarks/Tools/Specific)
   • Added Focus question (All/6 individual passes)
   • Added Severity filter question (All/High/Medium/Low)
   • Added Changed files filter question (All/Since date)
   • Added Jira epic question (None/Specify)

Documentation:

• Updated CLAUDE.md to correct skill count (10 total tracked skills)

Benefits

• Improved discoverability: Users see all available options with descriptions
• Better UX: No need to remember argument syntax or consult documentation
• Clearer defaults: Each question specifies a sensible default
• Flexible input: "Other" field support for custom values where needed
• Consistent interface: All updated skills now use the same interaction pattern

Testing

Tested activity-summary and open-prs skills with the new interactive prompts - both worked correctly.

🤖 Generated with Claude Code

[atheurer]

Nice improvement! Moving to interactive AskUserQuestion prompts makes the skills much more discoverable and user-friendly.

A few minor observations/suggestions for follow-up if you want to address them:

1. Residual CLI flag references:

   • activity-summary/SKILL.md: mentions In --all mode (could reference author_mode == "all" or "When author scope is All").
   • architecture-review/SKILL.md: Step 4 mentions When --scope limits... and When --repo is specified....
   • codebase-audit/SKILL.md: Step 4 mentions When --scope all and If --since is specified.
   • pr-review/SKILL.md: Step 8 mentions If --prior was provided.

2. new-repo details prompt:

   • Question 2 offers choices Tool, Benchmark, Custom (defaulting to Tool), but step 2 expects name|description from the Other field. Since name/description are mandatory for creating a repo, if a user simply selects Tool without typing into Other, parsing could fail. Splitting into separate questions (e.g. repo name and repo description) or refining the prompt instruction might be clearer.

3. CLAUDE.md table count:

   • The tree was updated to 10 skill directories, but the table in CLAUDE.md still has 11 entries because it lists debug-log (which does not exist in plugins/crucible-tools/skills/).

LGTM overall!

[k-rister]

Addressed all three follow-up suggestions in 6b0af7a:

1. Residual CLI flag references — replaced --all/--scope/--repo/--since/--prior (and a couple more of the same kind not explicitly called out: --severity, --epic, --focus in codebase-audit) with references to the actual question/variable names across activity-summary, architecture-review, codebase-audit, and pr-review.
2. new-repo details prompt — split the combined Tool/Benchmark/Custom + name|description pipe-parsing question into three separate questions (type, name, description), following the same None/Specify-style free-text pattern already used successfully elsewhere (open-prs, codebase-audit's Jira epic question). Name is now flagged as mandatory with a fallback to ask a direct follow-up if the user doesn't use the Other field.
3. CLAUDE.md skill count — removed the debug-log row from the table; confirmed via git ls-tree that only 10 skills are actually tracked in this repo.

🤖 Generated with Claude Code

[atheurer]

LGTM! The updates to interactive AskUserQuestion prompts look clean and well-structured across all skills.