feat: enhance workflow discipline, add community docs, and failure decision tree

[wangzhuo-coding]

Summary

This PR enhances the slide-alchemy project by incorporating best practices observed from the popular ppt-master repository, focusing on four areas:

1. SKILL.md Workflow Discipline Enhancement

• Added 9 Global Execution Rules with rationale for each rule
• Added GATE/BLOCKING/Checkpoint triple-layer protection to every workflow step
• Added Global Rule 8: Lead agent must re-read element_analysis.json after parallel page dispatch
• Made GATE conditions and Checkpoint artifacts explicit and verifiable for each of the 10 steps

2. Community Governance Documents

• CONTRIBUTING.md: Contribution guide with PR triage, SVG compatibility rules, and scope boundaries
• CODE_OF_CONDUCT.md: Contributor Covenant v2.0 with enforcement ladder

3. README Quality Improvements

• Added competitor comparison table (slide-alchemy vs ppt-master vs PPTAgent)
• Added honest expectation management section
• Enhanced installation instructions

4. Failure Decision Tree

• references/failure-decision-tree.md: Comprehensive remediation guide covering all failure modes
• Includes escalation path for unresolved issues

Files Changed

• skill/slide-alchemy/SKILL.md — Enhanced with execution discipline
• CONTRIBUTING.md — New
• CODE_OF_CONDUCT.md — New
• skill/slide-alchemy/references/failure-decision-tree.md — New
• README.zh-CN.md — Enhanced

[CodingFeng101]

Thanks for this PR. Overall, I think the direction is good, and many parts are worth keeping, especially:

• the gate / blocking / checkpoint discipline in SKILL.md
• re-reading element_analysis.json after parallel page work
• the expanded failure-decision-tree.md
• CONTRIBUTING.md / CODE_OF_CONDUCT.md
• the README improvements around comparison, honest expectations, unattended mode, and contribution entry points

However, I do not think this PR should be merged as-is yet. A few things need to be addressed first:

1. It currently conflicts with main

After #1 was merged, this PR now conflicts with the latest main in skill/slide-alchemy/SKILL.md.

Please rebase on the latest main or open a refreshed PR.

2. The README wording around “partial pages” should be more accurate

The PR currently says:

即使只请求部分页面，工作流仍会完整运行（包括所有页面的底图生成和组件分析），因为组件复用需要全局分析。

This is not quite accurate. I would suggest wording closer to:

如果只请求部分页面，工作流会针对所选页面完整运行；当需要跨页复用组件时，可能会参考相邻或相关页面。

In English, the intended meaning is:

If only selected pages are requested, the workflow should run fully for those selected pages. When cross-page component reuse is needed, adjacent or related pages may be referenced.

This avoids implying that the whole deck must always be processed even when the user only requests a subset.

3. Some fallbacks in failure-decision-tree.md should be tightened

Most of the expanded failure guidance is useful, but a few fallbacks conflict with the current quality rules. For example:

As a last resort, manually mask the remaining content with a solid fill matching the surrounding area.

The current workflow should avoid using white boxes or rough local masking as a substitute for clean base generation. I would suggest changing it to something like:

If repeated generation still leaves remnants, regenerate from a cleaner text description or split the base into simpler regions. Do not hide remnants with white boxes or local masking.

In other words: if the generated base still has leftover text/icons/cards, try a clearer base-generation description or split the complex base into simpler regions. Do not cover the leftovers with white boxes or local masks.

Similarly, text positioning and escalation guidance should stay as actionable agent repair steps instead of accepting approximate results too early or immediately turning failures into GitHub issues.

4. Please split this into smaller PRs

To make review and merging easier, I suggest splitting this into:

• PR A: SKILL.md execution discipline, including gate / blocking / checkpoint rules and re-reading element_analysis.json.
• PR B: expanded failure-decision-tree.md, with fallbacks tightened so they do not conflict with the current quality rules.
• PR C: README improvements, including comparison, honest expectations, unattended mode, contribution entry points, and the corrected partial-page wording.
• PR D: CONTRIBUTING.md / CODE_OF_CONDUCT.md, which can be reviewed and merged independently.

I am happy to merge the parts that are independent and low-risk first, such as CONTRIBUTING.md / CODE_OF_CONDUCT.md and the clearer README improvements. The SKILL.md and failure-decision-tree.md changes are valuable too, but they should be reviewed separately because they affect execution behavior.