skills/session-history/SKILL.md
Access session history to continue work, see what happened, or search past sessions. Use this instead of relying on memory - the raw data is always more accurate.
npx skillsauth add srobinson/nancy session-historyInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Core principle: Less context stuffing, more lookup. Your memory degrades (context rot), but the raw session data doesn't.
# Helper script location
SH=~/.claude/skills/session-history/session.sh
# What's the status? (current + previous summaries)
$SH status
# List all sessions for this project
$SH list
# Last N messages from current session (see what you ACTUALLY did)
$SH tail 10
# All user prompts from current session (intent trail)
$SH prompts
# Summaries from a session (auto-generated)
$SH summaries
# Search across all sessions
$SH grep "search term"
# Files touched in a session
$SH files
| Command | Description |
| ---------------- | ---------------------------------------------- |
| status | Overview: current + previous session summaries |
| list | All sessions with first prompt and summaries |
| current | Print current session ID |
| previous | Print previous session ID |
| tail [n] [id] | Last N messages (default: 10, current session) |
| prompts [id] | All user prompts from session |
| summaries [id] | Auto-generated summaries |
| grep <term> | Search across all project sessions |
| files [id] | Files touched (from tool calls) |
$SH status to see where things left off$SH tail 20 to see what you ACTUALLY did recently$SH prompts $(session.sh previous)$SH grep "feature name"Adhoc sessions (you just ran claude):
~/.claude/projects/<encoded-cwd>/Worker agents (HumanWork orchestrated):
HW_SESSION_DIR and HW_SESSION_ID~/.claude/projects/<encoded-cwd>/
├── <session-id>.jsonl ← each session
└── ...
JSONL contains:
- type: "user" → user prompts
- type: "assistant" → responses (with tool calls)
- type: "summary" → auto-generated summaries
- type: "file-history-snapshot" → file state
SH=~/.claude/skills/session-history/session.sh
# See what happened
$SH status
# Get the previous session's prompts
$SH prompts $($SH previous)
# See files that were modified
$SH files $($SH previous)
testing
Mark success criteria as complete in SPEC.md. Use when a criterion has been verified and should be checked off.
tools
Send a message to the orchestrator. Use for blockers, progress updates, requesting review, or communicating status during autonomous execution.
development
Create a task specification (SPEC.md) through interactive requirements elicitation. Use when helping users define what they want to build before autonomous execution begins.
testing
Check for orchestrator messages.