AGENT SKILL

spicy-remove

Remove egregious content from campaign story history and game state when model safety blocks cause empty responses or zero candidates

License 许可
MIT
最近更新
2026-11-14
Stars
★ 6
Forks
0

01 前言 Front Matter

name
spicy-remove
description
Remove egregious content from campaign story history and game state when model safety blocks cause empty responses or zero candidates

03 SKILL.md 正文

Spicy Remove Skill

Use this skill when a campaign's interaction streams fail with empty responses or zero candidates due to model safety blocks triggering on egregious content in the story history or game state.

Principle: Egregious vs General Spicy Content

This workflow is designed to only remove egregious safety-violating content that blocks model generation, while leaving general consensual sex or spicy scenes intact:

  • Egregious/Violating content to remove:
    • Non-consensual sexual acts, sexual assault, forced submission, or torture.
    • Incestuous themes or terminology.
    • Explicit/vulgar slang words (e.g., cumswap, cum-swapping, cum-streaked, blowjob, gangbang, deflower, prostitute).
  • General spicy content to keep:
    • Consensual romance, intimacy, passion, and generic spicy descriptions that do not involve non-consensual themes, violence, or explicit vulgar slang words.

Action: Clean Campaign In-Place in Firestore

Execute the utility script scripts/spicy_remove.py to recursively scan and sanitize the campaign's current_state and story documents.

Usage

# Run a dry-run to preview changes first
./vpython scripts/spicy_remove.py <campaign_id> --dry-run

# Apply the sanitization in-place in Firestore
./vpython scripts/spicy_remove.py <campaign_id>

What the Script Does:

  1. Automatically locates the campaign and its owner user UID across all users in Firestore.
  2. Recursively traverses all nested dictionary and list values in game_states/current_state (including character histories, log entries, and social_hp_challenge fields).
  3. Recursively scans all documents in the story subcollection.
  4. Identifies egregious keywords and cleanses slang words like cum (safely ignoring common words like document or cumulative).
  5. Replaces unsafe string values with safe, clinical descriptions:
    • Short fields/statuses: replaced with clean summaries (e.g., "Under guard, cooperating with the Shogunate.").
    • Narrative blocks: replaced with a safe story progression block describing the capture and confinement of captives under guard.
  6. Saves the updated fields back to Firestore.

04 同仓库更多 Skills