Case Writer Hybrid

Expand a structured brief in `content-production/inbox/` into a reusable long-form markdown article draft, then run a local writer / critic / judge quality l...

Security Scan

The skill is internally consistent with its description: it reads a local structured brief, runs a local writer/critic/judge loop, and writes draft and sidecar files to the repository — it requests no credentials, has no install step, and performs only local filesystem operations.

Purpose & Capability

Name/description match behavior: the skill parses brief markdown from content-production/inbox/, composes an article and writing pack, runs local quality rounds, and writes outputs under content-production/drafts (and a quality-gate path if needed). No unrelated capabilities (cloud access, external services, or unrelated system privileges) are requested.

Instruction Scope

SKILL.md and runtime instructions are narrowly scoped to reading the specified brief file, producing stage-1 drafts and sidecars, and running up to three local rounds of writer->critic->humanizer->judge. The instructions explicitly avoid downstream steps if quality fails. There are no instructions to read unrelated system files, environment variables, or to transmit data externally.

Install Mechanism

No install spec is provided (lowest-risk), but the skill includes runtime.py and depends on skill_runtime.writing_core. You should confirm how the repository is deployed and ensure required Python dependencies (e.g., skill_runtime) are from a trusted source. Overall no network download/install is embedded in the skill bundle itself.

Credentials

The skill declares no required environment variables, no credentials, and no config paths. Its filesystem access is proportional to its purpose (reading briefs and writing drafts inside content-production).

Persistence & Privilege

always is false and the skill does not request persistent or platform-wide privileges. It does not modify other skills' configs. Autonomous invocation is allowed by default but is not combined with other concerning privileges.

Assessment

This skill appears coherent and limited to local drafting work. Before installing, verify these practical points: (1) Inspect or confirm the provenance of the skill_runtime.writing_core module and any requirements.txt used at runtime — that module does the real work and should be trusted. (2) Run the skill in a sandboxed environment (virtualenv/container) so its filesystem writes are contained. (3) Ensure briefs you place under content-production/inbox/ do not contain secrets (credentials, private keys, or sensitive PII) because the skill reads and writes those files locally. (4) If you plan to chain downstream skills (generate-image, wechat-formatter), audit those skills separately for network access or credential requests. (5) If you need higher assurance, review the rest of runtime.py (and skill_runtime.writing_core) end-to-end to confirm there are no hidden network calls or logging to external endpoints.

Turn the structured brief into a stage-1 article draft, a writing pack sidecar, and a review trace that can flow into image generation and WeChat formatting.

Quick Start

Run the default command:

.venv/bin/python -m skill_runtime.cli run-skill case-writer-hybrid --input content-production/inbox/20260403-ai-content-system-brief.md

Prepare Input

Expect a markdown file with these sections:

• 基础信息
• 核心观点
• 背景与语境
• 论证方向
• 可用案例 / 素材

Important fields inside 基础信息:

• date
• slug
• topic
• target_reader
• publish_goal

Follow Drafting Workflow

1. Read the brief and preserve explicit user-supplied claims, framing, and evidence.
2. Use 论证方向 as the backbone for the major argument sections.
3. Pull material from 可用案例 / 素材 into the matching sections instead of inventing new examples first.
4. Build a stable first draft structure: title, 导语, 问题提出, 核心判断, 论证段, 结论, 可传播总结.
5. Run up to 3 local rounds of writer -> critic -> humanizer-zh -> judge.
6. If the score still fails after 3 rounds, stop and emit a quality-gate notice instead of continuing downstream.

Write Output

Write the markdown draft to:

content-production/drafts/<slug>-article.md

Also write:

content-production/drafts/<slug>-writing-pack.md
content-production/drafts/<slug>-writing-pack.json
content-production/drafts/<slug>-review-trace.json

If the draft fails the quality gate after 3 rounds, also write:

content-production/published/YYYYMMDD-<slug>-quality-gate.md

Respect Constraints

• Treat the result as a controlled local draft loop, not a free-form creative writer
• If the brief is sparse, preserve structure and fill gaps conservatively
• Prefer explicit user-supplied arguments over invented framing
• Do not continue to image generation or formatting if the quality gate fails

Read Related Files

• Runtime entry: skill_runtime/engine.py
• Usage guide: docs/case-writer-hybrid-execution-spec.md
• Downstream skills: generate-image, wechat-formatter