Diagnose and reduce noise in agent-user conversation transcripts.
npx skills add https://github.com/MattWenJun/ClaWiser --skill noise-reduction
Please help me install this Agent Skill. Skill: noise-reduction Source: https://github.com/MattWenJun/ClaWiser/tree/main/assets/noise-reduction Command: npx skills add https://github.com/MattWenJun/ClaWiser --skill noise-reduction
Noise in conversation transcripts degrades memory recall and makes it hard to retrieve relevant context. This Skill provides structured noise profiling, metadata stripping, and conservative filtering to preserve meaningful user-agent content while removing noise patterns.
Run diagnose-noise.js <YYYY-MM-DD> to generate an initial noise profile for that date.
High-intent search queries and answers about installing and using this skill.
To reduce conversation noise, this Skill profiles transcript patterns and applies rule-based filtering to strip metadata wrappers, preserving meaningful content for better memory recall.
Yes, noise profiling works across OpenClaw sessions and transcripts from channels like Telegram, Discord, and Slack, filtering irrelevant messages to improve transcript quality.
You diagnose noise patterns by running the validation script diagnose-noise.js with a target date, generating a structured noise profile that catalogs irrelevant content.
The best way to strip metadata wrappers safely is using conservative rule-based filtering, which removes irrelevant noise patterns while implementing safeguards to preserve essential context.
Noise reduction includes validation and reporting scripts that explicitly quantify compression, false positives, and missed noise, ensuring false positives are tracked for continuous improvement.
You should avoid rule-based transcript filtering when conversations contain unstructured noise patterns that fall outside the diagnosed profile, potentially triggering conservative safeguards and missing noise.
zjykzkIngest documents, text, and URLs into interconnected Obsidian wiki pages.
0xBigBossExtract Claude Code and Codex session JSONL transcripts into Markdown.
harukainguyen1411Consolidate session memory and state before context compression.
anserIndicusFilter community data by removing off-topic content, spam, and bot responses.
