- Home
- Skills
- Prompt & Skill Engineering
- claude pickup
claude pickup
Context loss is the primary failure mode of long-running AI development sessions.
Free
Works with the AI Agents you already use
claude pickup
Example session with this skill installed
/claude-pickup
- Read your context and instructions
- Compiled the claude pickup
I have located the SESSION.md file and verified it against the current git state. I've processed the objective to refactor the authentication middleware and reconciled the pending tasks against the 3 commits that landed since the last handoff.
Loaded agensi-cli @ feature/auth-fix — handoff from 2025-05-14 14:20:00Z (2 hours ago) — with uncommitted changes the handoff doesn't mention.
Next steps
- Ask me to list the reconciled open tasks to see what is actually remaining.
- Direct me to start on the next task using the uncommitted changes as a starting point.
- Query the archive if you need to know why the previous session abandoned the JWT approach.
Connects securely to your tools. The creator never sees your data.
About this skill
Context loss is the primary failure mode of long-running AI development sessions. When an agent loses its history, you waste tokens and time re-explaining the same architectural decisions or watching the agent walk back into a dead end the previous session already solved. This skill provides a specialized protocol for resuming work with surgical precision.
What it does
- Session Restoration reads the
SESSION.mdhandoff file to recover goals, hard-earned decisions, and known dead ends that aren't visible in git history. - State Reconciliation verifies the handoff against the current repo state, identifying commits or manual changes that landed after the last session ended.
- Drift Detection checks branches, stashes, and dirty files to ensure the agent is working on the correct line of development.
- Narrow Enrichment loads only the specific files required for the immediate next step, preventing context window saturation.
- Archive Searching provides a targeted grep-based mechanism for querying historical session data without loading massive archive files.
How it works
- Locate and Read the agent finds the handoff file and performs a full read to capture the "map" of the current project status.
- Verify Reality it runs git commands to compare the handoff's timestamped claims against the actual branch history and uncommitted diffs.
- Reconcile Tasks it filters the "Next Steps" from the handoff against the repo state to determine which items are truly open, closed, or dormant.
- Minimal Load the agent performs existence and symbol checks on a maximum of 6 relevant files.
- Quiet Ready it outputs a single status line and stops, waiting for your instruction rather than generating redundant reports.
Frameworks & tools
This skill utilizes standard Unix utilities and Git including ls, grep, sed, wc, git status, git log, git diff, and git show. It is designed to work in any terminal-capable agent environment like Claude Code or Cursor.
Why this beats prompting it yourself
Generic prompts often lead to "over-loading" where an agent reads the entire repo and burns 30% of your context window before starting work. This skill enforces a strict 15% context budget and prevents the agent from re-running tests or builds during the orientation phase, saving you both tokens and time.
Use cases
- Multi-day projects where you need to resume complex coding tasks across different AI sessions.
- Context recovery after a model hits its token limit and requires a fresh start.
- Handoff verification when you have manually committed code between AI sessions and need the agent to catch up.
- Troubleshooting by querying
SESSION_ARCHIVE.mdto find out why a specific architectural path was previously abandoned.
Known limitations
Does not support SESSION.md files over 600 lines without two-pass reading. Requires a repo with git initialized and a valid SESSION.md file created by a compatible handoff tool.
Requires existing SESSION.md file. Only reads up to 6 files during enrichment. Max 500 lines per file for full reads. Archive files are search-only.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed today
- Free with an Agensi account
Requires