About this skill

Problem

After long sessions, resumed tasks, or tool-heavy turns, it is easy to lose a traceable record of what the user actually asked and what the assistant actually replied. This skill turns the current thread into a project-local Markdown log, avoiding reliance on chat screenshots, manual copy-paste, or visible docs mixed with hidden system prompts, tool schemas, and chain-of-thought content.

How It Works

The skill tracks only the current conversation and is enabled on demand. It creates YYYY-MM-DD-HHMMSS-conversation.md in the project root, then appends one completed logged turn at a time: the timestamp for the user message, the user prompt text, and a one-sentence summary of the assistant reply. It prefers a project-local Stop hook in .claude/settings.local.json so logging runs automatically at assistant stop points; if persistent hooks are unavailable, it falls back to explicit in-conversation appends. Session bookkeeping lives in .claude/log-current-conversation/ to avoid duplicate entries.

Boundaries

It does not backfill older threads or record hidden prompts, developer prompts, tool schemas, or chain-of-thought. If the log file is missing but its path is known, it recreates the same file; if the path cannot be recovered, it starts a new timestamped file. It fits project-local conversation summaries and audit trails, not global chat exports or multi-session logging management.

Screenshots

Use Cases

While debugging an API error, write the current thread's prompts and assistant conclusions into a project Markdown log turn by turn.
When resuming the same task across days, continue appending to the known current-session log without backfilling older threads.
Before a code review, consolidate the current conversation's questions, requests, and reply summaries into a single project file.
After multiple agent tool calls, record only user-visible reply summaries without writing hidden tool schemas into the log.

Best For

Claude-based debugging engineers who need a project Markdown trace of current-session prompts and reply summaries.
Agent workflow maintainers who need a single project-local log for the current thread without older sessions.
Developers handing off tasks who need key requests and assistant conclusions from the current session.
Engineers auditing AI collaboration who need user-visible prompts and replies without hidden prompt content.