Feasibility Ai Offline Writer

Quick Workflow

1. Pick the workflow family

   • Chapter drafting -> read references/system-prompts.md and fetch the relevant chapter prompt with scripts/lookup_prompt.py
   • QA / review / revise / extract / consistency -> read references/task-prompts.md or references/review-prompts.md
   • Scheme building -> read references/scheme-prompts.md
   • Industry expert alignment -> read references/expert-index.md
   • Outline selection -> read references/outline-index.md and use scripts/select_outline.py

2. Inspect exact prompt entries

   • python scripts/lookup_prompt.py --list-groups
   • python scripts/lookup_prompt.py --group system
   • python scripts/lookup_prompt.py --key system/government
   • python scripts/lookup_prompt.py --search consistency

3. Inspect outline coverage

   • python scripts/select_outline.py --list-majors
   • python scripts/select_outline.py --major A1 --list-subgroups
   • python scripts/select_outline.py --major A1 --subgroup municipal
   • python scripts/select_outline.py --search sewage

Bundled Data

• assets/prompts.json ? full recovered prompt/config corpus
• assets/outlines.json ? full recovered outline/template corpus

References

• references/prompt-taxonomy.md ? prompt families and usage map
• references/system-prompts.md ? system-level writing roles and output rules
• references/task-prompts.md ? generic task prompts for write/review/revise/extract/polish
• references/review-prompts.md ? stricter review-agent prompts
• references/scheme-prompts.md ? interviewer / architect / reviewer trio
• references/expert-index.md ? expert-category map and lookup guidance
• references/outline-index.md ? major/subgroup outline catalog
• references/project-structure.md ? recovered project file structure and local data model

Practical Guidance

• Treat the bundled prompts as prompt resources, not as claims that a specific hosted model is required.
• Prefer reading only the reference file relevant to the current job; use the scripts for exact entry lookup.
• When drafting content, combine:
  1. one system role,
  2. one chapter/task prompt,
  3. optional expert prompt,
  4. optional outline/template constraints.
• When auditing content, prefer review/* over tasks/* if we want stricter audit framing.

Common Patterns

Draft a chapter

• Read references/system-prompts.md
• Fetch chapter config like government_2023/chapters/4_site
• Optionally fetch experts/<category>
• Build the final prompt in your working context

Review and revise a chapter

• Start with review/review or tasks/review
• Then use review/revise or tasks/revise
• Use tasks/extract / review/extract when we need structured facts first

Build a scheme interview flow

• Read references/scheme-prompts.md
• Use scheme/interviewer to ask for missing data
• Use scheme/architect to transform facts into structured patches
• Use scheme/reviewer to check missing technical fields and risks

Notes

• This skill is derived from the locally recovered FeasibilityAI desktop bundle.
• The recovered client labels the upstream provider as deepseek and the default model as deepseek-chat, but this skill is intentionally offline and model-agnostic.