Why not just use Claude /compact, plan.md, or resume from summary?
Use them when you can. They are good Claude-native workflows if you catch context pressure early and want to stay inside Claude.
The limit is the boundary. /compact and /resume do not create a neutral packet for Cursor, Codex, Gemini, or another AI tool. ShardStitch reads local project state, separates verified disk facts from inferred claims, and formats the packet for the next tool.
How is ShardStitch different from a /handoff skill?
/handoff helps when you remember to run it inside one tool. It is useful, but it still depends on the current tool and the current session producing the handoff.
ShardStitch builds a local, disk-aware handoff from the project itself: git diff, changed files, plans, notes, failed attempts, verification state, and the next action. Then it formats that packet for the next AI tool, even if the old session is already too bloated, blocked, or dead.
What if I watch my usage window and avoid big tasks near the limit?
That is a smart workflow. Treating the usage window like a budget helps you avoid predictable cutoffs.
ShardStitch is for the cases where planning was not enough: the task ran long, the session became bloated, the model became unavailable, or the work is already mid-flight and waiting for reset would break momentum.
Does ShardStitch need the old AI session alive?
No. It reads what survived locally: git diff, changed files, recent commits, dependency graph, project notes, local tool logs, and persistent project memory.
That is the core difference. Summary tools ask the old session to explain itself. ShardStitch reads the project state from disk.
Is this a replacement for good notes?
No. Good notes still help. A maintained plan.md can make any recovery cleaner.
ShardStitch exists because notes drift, sessions die, and humans forget to prepare. It uses notes when they exist, but it does not depend on perfect notes being maintained in advance.
What does a good handoff include?
A reliable handoff is a forward brief, not a backward summary. It should capture the goal, current state, touched files, decisions, failed attempts, verification status, and the next action.
ShardStitch separates verified disk facts from inferred next steps, so the next AI does not confuse guesses with truth. The handoff becomes the next session's ground truth: lean, verified, and tied to what actually changed on disk.
When is ShardStitch most useful?
- The session is already rate-limited or dead.
- The chat is too bloated to trust its own summary.
- You want to move from Claude to Cursor, Codex, Gemini, or another tool.
- You need git state, changed files, dependency risk, and intent in one handoff.
- You want a clean restart without dragging every old debugging branch forward.
Does my code leave my machine?
Not during normal recovery. ShardStitch recovery, dashboard, CLI, MCP, memory, and extension workflows are local.
License validation goes to Polar at activation and during quiet periodic re-checks. That request does not include your code, conversations, git diff, notes, or handoff content.
Which tools are supported?
ShardStitch formats handoffs for 31 AI handoff targets across native integrations and web-chat handoff, including Claude Code, Codex, Gemini CLI, Grok, Cursor, Devin, Aider, DeepSeek, Claude Desktop, OpenCode, Trae, Kiro, Amazon Q, Cline, Roo Code (archived May 2026), Kimi, Qwen Code, Antigravity, and more.
Is this a subscription?
No. ShardStitch is $49 as a one-time lifetime license, with no subscription or recurring fees.