ship the /ask skill

[acunniffe]

Extending on @jwiegley's search + continue work, this PR introduces a skill that lets any agent read previous prompts for any LoC it's looking at.

It's useful when trying to figure out why AI-code is the way it is and I've found it very useful during planning. I though there'd be some crazy work required to make agents call it, but this seems to do the trick:

Claude.md|AGENTS.md

- In plan mode, always use the /ask skill so you can read the code and the original prompts that generated it. Intent will help you write a better plan

Also included
[ ] updates to VSCode plugin empty prompt state
[ ] reading prompts for logged in teams from CaS Cache, then CaS, then Local DB, then notes
[ ] CaS read cache

Note: This should ship with a VSCode update, but they are technically independent and won't break if out of sync

Separate PR with big doc updates + readme updates in the works

---

[git-ai-cloud-dev]

Stats powered by Git AI

🧠 you    ██░░░░░░░░░░░░░░░░░░  10%
🤖 ai     ░░██████████████████  90%

More stats

• 1.3 lines generated for every 1 accepted
• 0 seconds waiting for AI
• Top model: claude::claude-opus-4-6 (727 accepted lines, 950 generated lines)

AI code tracked with git-ai

[devin-ai-integration]

Devin Review found 1 new potential issue.

View 20 additional findings in Devin Review.

[devin-ai-integration]

🟡 CAS fetch concurrency cap silently drops all prompts beyond the first 3

triggerCASFetches is called once when a blame result arrives. It slices to the first 3 prompts (promptsToFetch.slice(0, 3)) and starts fetching them. When each fetch completes, the .then() callback at blame-lens-manager.ts:1552-1571 only calls updateStatusBar — it never re-invokes triggerCASFetches to start fetching the next batch.

Root Cause and Impact

The comment says "Cap concurrent fetches at 3" (line 1546), implying a concurrency limit. But because triggerCASFetches is only called once per blame result (at blame-lens-manager.ts:523 and blame-lens-manager.ts:601), and the .then() completion handler does not call it again, the slice(0, 3) acts as a hard total limit, not a concurrency limit.

If a file has, say, 8 distinct AI prompts with messages_url but no messages, only the first 3 will ever be fetched. The remaining 5 prompts will permanently show "Loading prompt from cloud..." in the hover tooltip (as coded at blame-lens-manager.ts:1216-1218) until the user triggers a full blame re-fetch (e.g., by saving the file).

The fix is to call this.triggerCASFetches(blameResult, ...) inside the .then() callback after a successful fetch, so the next batch of up to 3 is started once a slot frees up.

Prompt for agents

In agent-support/vscode/src/blame-lens-manager.ts, inside the triggerCASFetches method (around line 1531), the .then() callback at line 1552 should re-invoke triggerCASFetches after a successful fetch completes, so that the remaining prompts beyond the first 3 are eventually fetched. Specifically, after the updateStatusBar call at line 1559, add a call like: this.triggerCASFetches(blameResult, documentUri); You will need to capture documentUri in the closure scope. The triggerCASFetches method already filters out prompts that are in casFetchInProgress or already have messages, so re-invoking it is safe and will only start new fetches for the remaining prompts.

---

Was this helpful? React with 👍 or 👎 to provide feedback.