Extract and repair valid JSON from malformed language model outputs.
npx skills add https://github.com/JGalego/Bag-of-Tricks --skill salvage-jgalego
Please help me install this Agent Skill. Skill: salvage Source: https://github.com/JGalego/Bag-of-Tricks/tree/main/salvage Command: npx skills add https://github.com/JGalego/Bag-of-Tricks --skill salvage-jgalego
This Skill addresses the issue of extracting and repairing valid JSON from the verbose, often malformed output generated by language models, ensuring that downstream systems receive structured data.
Use the salvage skill to extract and repair JSON from the model's response: salvage "Here's the model's response with malformed JSON."
High-intent search queries and answers about installing and using this skill.
Extracting valid JSON from chatty LLM output involves filtering the response through a stdin-to-stdout pipe that removes markdown fences and repairs trailing commas. This ensures downstream systems receive clean, structured data without manual edits.
Fixing malformed JSON with Python literals and smart quotes requires a data repair filter that normalizes these characters into standard JSON syntax. The salvage tool operates as a CLI filter to automatically correct these common LLM formatting issues.
No additional dependencies are required to parse and repair JSON from CLI tool output. The salvage skill operates independently as a stdin-to-stdout filter, ensuring you can clean LLM responses without adding extra packages to your environment.
Yes, you can use a stdin-to-stdout filter to clean verbose LLM responses for data processing. The salvage skill intercepts chatty model output, strips non-JSON text, and outputs repaired JSON directly to standard output for immediate downstream consumption.
JSON parsing fails on language model output because conversational text, markdown fences, and Python-style literals violate strict JSON syntax rules. Data repair tools isolate the JSON structure and correct these formatting deviations to restore validity.
lrhodinPipe arbitrary text through an LLM to generate concise responses.
yangchuanshengQuery the World Bank indicators API and extract newest non-null values.
73aiSanitize model responses through a Model Armor template.
sungaoxiang-backendGenerate pre-trial evidence collection plans with structured checklists and JSON-like outputs.
