← Back to Skills Marketplace
Session Feedback Analyzer
by
_silhouette
· GitHub ↗
· v1.0.1
· MIT-0
107
Downloads
0
Stars
1
Active Installs
2
Versions
Install in OpenClaw
/install session-feedback-analyzer
Description
Parse Claude Code session JSONL to extract implicit user feedback signals. Detects skill invocations (tool_use blocks with name="Skill" or /slash-commands),...
Usage Guidance
This skill appears to do what it says: it parses local Claude session JSONL files to emit feedback events and compute per-skill metrics. Before installing/running, consider: 1) it reads conversation files under ~/.claude/projects (sensitive data) — run with --no-snippets if you want to avoid storing message text; 2) the scripts expect to import lib.common from a parent repo root that is not included in the package, so you may need to run the tool from the larger repo or supply the missing helper functions; 3) run the bundled tests (pytest) in a safe environment to ensure the script runs correctly; 4) inspect where output (feedback.jsonl) will be written and treat that file as sensitive; 5) because there is no install step, no external code is fetched, but exercise caution when running any script that reads local data.
Capability Analysis
Type: OpenClaw Skill
Name: session-feedback-analyzer
Version: 1.0.1
The session-feedback-analyzer is a legitimate utility designed to parse local Claude Code session logs (~/.claude/projects) to compute skill performance metrics. The Python scripts (analyze.py and metrics.py) implement keyword-based heuristics to classify user feedback without any network activity, unauthorized data exfiltration, or suspicious execution patterns. The bundle includes robust privacy controls, such as an opt-out configuration check and a flag to suppress user message snippets, and its behavior aligns strictly with its documented purpose.
Capability Assessment
Purpose & Capability
Name/description match the behavior: the scripts parse ~/.claude/projects JSONL, detect skill invocations, classify user responses, and compute metrics. The requested resources (local session files) are proportional to the stated goal. Minor incoherence: the scripts import lib.common from a parent repo root (sys.path manipulation), but lib/common is not included in the bundle — this indicates the code expects to run inside a larger repo layout and may fail when run standalone.
Instruction Scope
SKILL.md instructs running python3 scripts/analyze.py against a session directory and to write feedback.jsonl. The instructions only reference local session files and the analyzer output; they do not send data externally or request unrelated system credentials. Privacy note: the analyzer reads user message text from session files (it offers --no-snippets to strip text), so running it will access potentially sensitive conversation content.
Install Mechanism
There is no install specification (instruction-only), so nothing will be downloaded or installed automatically. The code requires a Python runtime and expects to be executed from a repo layout that provides lib.common; that dependency is not bundled here and could cause runtime errors. No network downloads or external package installs are requested.
Credentials
The skill requires no environment variables, binaries, or external credentials. It accesses the filesystem (session JSONL under ~/.claude/projects by default), which is appropriate for a session-log analyzer but should be considered sensitive. The code may write output files (feedback.jsonl) to the current directory or a provided output path.
Persistence & Privilege
The skill does not request persistent/always-on privileges. Flags show always:false and no special persistence. It writes its own output file when run, which is expected behavior for a CLI analyzer.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install session-feedback-analyzer - After installation, invoke the skill by name or use
/session-feedback-analyzer - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
v2.1: 13/13 POWERFUL at 92.2%, enriched SKILL.md docs, README added, example tag fix
v1.0.0
Session Feedback Analyzer 1.0.0 – Initial Release
- Extracts implicit user feedback from Claude Code session JSONL, classifying corrections, acceptances, and partials within a 3-turn influence window.
- Computes per-skill correction rates and generates structured feedback.jsonl for improvement pipelines.
- Supports dimension attribution (accuracy, coverage, reliability, efficiency, security, trigger_quality) for each correction event.
- Includes trend analysis, privacy controls, and a CLI/API for batch statistics and trend detection.
- Not for synthetic task suite evaluation; complementary to improvement-evaluator and improvement-learner.
Metadata
Frequently Asked Questions
What is Session Feedback Analyzer?
Parse Claude Code session JSONL to extract implicit user feedback signals. Detects skill invocations (tool_use blocks with name="Skill" or /slash-commands),... It is an AI Agent Skill for Claude Code / OpenClaw, with 107 downloads so far.
How do I install Session Feedback Analyzer?
Run "/install session-feedback-analyzer" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Session Feedback Analyzer free?
Yes, Session Feedback Analyzer is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Session Feedback Analyzer support?
Session Feedback Analyzer is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Session Feedback Analyzer?
It is built and maintained by _silhouette (@lanyasheng); the current version is v1.0.1.
More Skills