- Home
- Skills
- Agents & Orchestration
- bix-claude-skills: Load Session
Works with the AI Agents you already use
bix-claude-skills: Load Session
Load and summarize previous Claude Code sessions into your current context with smart templates and token estimation.
Free
bix-claude-skills: Load Session
Example session with this skill installed
/load-session payments-refactor
Or in plain language: "Load my payments-refactor session from yesterday as a handoff summary."
- Read your context and instructions
- Compiled the bix-claude-
Session: payments-refactor
Messages: 64 user + 71 assistant | Duration: 148 min | Size: 2.3 MB
1. Compact summary (recommended): ~500-1000 tokens
2. Full transcript: ~61,000 tokens (warning: uses 31% of context window)
3. Cancel
Loaded session: payments-refactor (compact: Handoff)
Subjective: What the user wanted
- Replace the hand-rolled retry loop in
src/billing/charge.tswith the sharedwithRetryhelper. - Keep the public
chargeCustomer()signature unchanged because the mobile app depends on it. - Familiar with the codebase; wanted minimal diffs.
Objective: What actually happened
- Edited
src/billing/charge.tsandsrc/lib/retry.ts; addedtests/billing/charge.retry.test.ts. - Ran
npm test: 212 passed, 2 failed incharge.retry.test.ts(timeout at 5000 ms). - Hit error
ERR_IDEMPOTENCY_KEY_REUSEDwhen the retry reused the same key.
Assessment: Analysis and conclusions
- The retry helper must generate a fresh idempotency key per attempt only for network errors, not for card declines.
- Decision: keep exponential backoff but cap at 3 attempts.
- Risk: webhook handler
src/billing/webhooks.tsstill assumes one attempt per charge.
Plan: What comes next
- Fix the 2 failing tests by faking timers in
charge.retry.test.ts. - Add the idempotency key rule to
withRetryoptions. - Review
webhooks.tsfor duplicate-event handling. - Open question: should the attempt cap be configurable per merchant?
Session loaded: payments-refactor
Mode: compact (Handoff)
Estimated tokens added: ~650
Next steps
- Ask Claude to continue with the two failing tests in
charge.retry.test.ts. - If you need exact wording from a message, run the skill again and choose full mode.
- Try the Decisions template if you only need the reasoning behind the retry cap.
Connects securely to your tools. The creator never sees your data.
About this skill
By default claude code's /resume only allows you to re-open a conversation from the same project, but there are many instances where you'd like to use the same conversation accross multiple related projects you may be working on. This skill bridges that gap by allowing you to inject previous Claude Code sessions directly into your current context without manual copy-pasting or context-window bloat.
What it does
- Session browsing lists historical sessions with metadata like project name, branch, and message count.
- Cost estimation calculates token usage and context window impact before you load data.
- Smart summarization uses specialized templates to extract architectural decisions, action items, or handoff notes.
- Sub-agent delegation processes massive transcripts through a dedicated summarizer to keep your main context lean.
- Transcript injection loads the full conversation history when you need every line of code and reasoning.
How it works
- Browse or specify a session by name or UUID to resolve the conversation history.
- Evaluate impact by reviewing the estimated token count and context percentage.
- Select a mode between a compact summary (using templates like "Decisions" or "Timeline") or a full transcript.
- Automatic cleanup ensures temporary transcript files are deleted immediately after the context is injected.
Frameworks & tools
Built for Claude Code environments. Requires Python 3 for session parsing and metadata extraction.
Why this beats prompting it yourself
Manual summaries require you to read thousands of lines of JSONL logs yourself. This skill automates the extraction, applies "Chain-of-Density" summarization via a sub-agent, and prevents the "current session loop" error that happens when manually feeding logs back to an agent.
Use cases
- Task resumption to pick up exactly where a previous session left off on a specific branch.
- Architectural review to extract only the trade-offs and decisions made during a long refactoring session.
- Context handoff to generate a clean summary of work done for another developer or a different agent.
- Code forensic to find the specific reasoning behind a past change without scrolling through terminal history.
- Claude Code only. It depends on the
~/.claude/projectssession layout. - Requires Python 3 (
python3orpython). It stops if neither is found. - It cannot load the current session into itself.
- Only user text, assistant text and completed sub-agent results are kept. Tool calls and tool results are dropped from the transcript.
- Sub-agent results in a transcript are cut at 2,000 characters each.
- Full mode puts the whole transcript in context. Large sessions can use a big share of the window, so check the estimate.
- Token counts are estimates (about 4 characters per token).
- Compact mode needs a sub-agent. On a standalone install it runs as a general-purpose Sonnet sub-agent using the bundled summarizer instructions, and temp-file cleanup is done by the skill instead of a hook.
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
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Permissions
File Scopes
The skill needs to read your past session files to list and load them. It writes only a temporary transcript, then deletes it. It does not modify, move or delete your session files. It does not read other parts of `~/.claude`.