Fireworks Skill Memory

Trust Score 56/100

Provides persistent, per-skill experience memory for Claude Code and Codex, injecting past lessons and distilling new ones across sessions.

triggers:skill memorycross-session learningdistill lessonsinject knowledgefireworks-skill-memory

GitHub SKILL.md

What it does

Fireworks Skill Memory gives Claude Code and Codex skills durable, skill-scoped memory across sessions. It injects past experience before a skill runs, captures error seeds and session notes during execution, and distills short lessons into KNOWLEDGE.md so future invocations learn from prior runs. Concrete use case: reduce repeated mistakes (wrong params, failed sequences) by surfacing past fixes automatically.

When to use it

Invoke this when you want an agent-driven long-term memory for coding workflows: installing or configuring skill memory, distilling lessons after a debugging session, or when repeated errors indicate a need for persistent knowledge. It triggers on requests about skill memory, cross-session learning, or installing/configuring fireworks-skill-memory.

What's included

Scripts: installer and CLI helpers referenced in the README; runtime hooks for Claude Code and Codex (pre/post/stop hooks described in the SKILL.md).
References: documentation links and example commands in the repo README.
Instructions: how to install (curl installer, npx skills add), CLI commands (inject, checkpoint, flush), and configuration environment variables.

Compatible agents

Best suited for Claude Code and Codex runtimes; can work with other agent runtimes that support pre/post hooks and CLI-driven installers.

Audit Summary

Persistent cross-session memory skill for Claude Code and Codex. Installs via curl|bash and sets up hooks that inject past experience and distill new lessons. Only 1 of 4 scripts (error-seed-capture.py) runs standalone — the other 3 depend on a memory_core package that isn't bundled. Security concerns include curl|bash installation and an auto-update mechanism that runs git fetch without user confirmation.

Watch Out

Installs via curl | bash — classic supply-chain risk vector
Auto-update mechanism (git fetch origin) runs without explicit user confirmation
3 of 4 scripts fail without the memory_core package (not bundled in scripts export)
Distiller backend sends transcript snippets to external APIs (Claude/OpenAI) by design

Missing Dependencies

memory_core (bundled module not included in scripts export)

Notes

Curl|bash installer and auto-update are the main security concerns. Scripts are well-documented and the architecture is sound, but the dependency on memory_core means scripts can't be evaluated in isolation. The skill solves a real problem for agent memory persistence.

Information

Repository: fireworks-skill-memory
Stars: 93
Installs: 0

Trust Score

Overall56

Security47

Code Quality62

Architecture72

Usefulness75

Related Skills

Caveman Compress

Compress natural-language memory files into a compact 'caveman' format to save LLM tokens while preserving code, links, and structure.

Sandbox0 Integration

Guidance and templates for integrating Sandbox0 sandboxing into AI agents — CLI/SDK patterns, templates, volumes, network policy, and deployment choices.

Codex Collab

Bridge Claude and Codex to run tasks, perform code reviews, and run background jobs for research or implementation workflows.

Smalltalk MCP

Connects an LLM (Claude/OpenAI/etc.) to a live Smalltalk (Squeak/Cuis) image over MCP for browsing, evaluating and modifying Smalltalk code.

Cross-Project Analytics

Query local, privacy-safe cross-project analytics to report on agent, skill, hook, and team performance; replay sessions and estimate token costs.

Go Data Structures

Authoritative guidance on choosing and using Go built-in and standard-library data structures, with practical best practices for slices, maps, arrays, container

Claude Code (Clawdbot)

Wrapper to run the local Claude Code CLI in headless or interactive (tmux) modes for code analysis, refactors, tests, and structured output.

Anthropic Security Basics

Practical security practices for Anthropic Claude integrations: API key management, input validation, prompt-injection defenses, and output scanning.

Back to Skills