Fix OTP generation, add graceful shutdown, and fix N+1 query

The task is "Wake up and do your best." This is a general directive to be proactive and do useful work. Let me look at the current state of the relentlessworks organization and see what needs attention.

Let me start by checking the organization's repositories, any open issues, any failing CI, etc.

All CI workflows are passing across all four repos. No open issues. The last task was fixing the deploy-pages workflow. Everything looks healthy.

Now, the task is "Wake up and do your best." This is a general directive. Let me think about what would be the most valuable thing to do:

1. All repos have passing CI
2. No open issues
3. The org has 4 repos: linksmith, hookrelay, notable, taskpilot, and .github

Let me think about what "do your best" means in this context. I should be proactive. Let me check if there are any improvements I can make. Let me look at the repos more closely - maybe I should:

• Check if any repos are missing topics (hookrelay has no topics)
• Check if any repos need README improvements
• Look at the code quality
• Maybe create a new service

Let me check hookrelay's topics - it has empty topics while others have good ones. That's a quick win.

Also, let me check the .github repo for organization-level configurations.

Let me also check if there are any open PRs across repos.

Let me be strategic. The most impactful thing I can do is:

1. Add missing topics to hookrelay (quick win)
2. Check the code in the repos for any issues or improvements
3. Maybe create a new agentic-first service to expand the portfolio

Let me start with the quick wins and then move to something more substantial. Let me check hookrelay's topics and add appropriate ones, and also check for open PRs.

Actually, let me think about what would be most valuable. The org builds agentic-first services. We have:

• linksmith (link shortener)
• notable (notes/knowledge base)
• taskpilot (task management)
• hookrelay (webhook relay)

A natural next service could be something like:

• A notification hub
• A contact/CRM service
• A bookmark/pinboard service
• A status page service
• A poll/voting service

But first, let me handle the quick wins: add topics to hookrelay, check for any issues in the code, and then decide on a bigger task.

Let me check hookrelay topics and also look at the repos' file structures to understand what's there.

Now I have a good understanding of all four repos. Let me look at the store files for the other repos to understand the patterns and find any issues.

Let me look at the store files for taskpilot, notable, and linksmith to see if there are any issues.

Now I have a thorough understanding of all four repos. Let me identify the key issues and improvement opportunities:

1. hookrelay - N+1 query problem in handleListEndpoints: The handleListEndpoints function calls s.store.ListDeliveries(ep.Handle, workspace, 10000) for each endpoint in a loop. This is an N+1 query problem that loads ALL deliveries for each endpoint just to count them. Same issue in handleGetEndpoint. This is a performance issue.

2. hookrelay - handleGetEndpoint also has the N+1 issue: It loads all deliveries (limit 10000) just to count them.

3. All repos - No go.sum file: The Dockerfile has COPY go.mod go.sum* ./ with a wildcard, so it handles the missing go.sum gracefully. But since there are no external dependencies, this is fine.

4. All repos - Missing .well-known/agent.md as a separate file: The help text is served at /.well-known/agent.md but it's the same as /help. This is fine per the spec.

5. All repos - No graceful shutdown: The main.go files use http.ListenAndServe which doesn't handle graceful shutdown. This means in-flight requests are dropped on SIGTERM. For a production service, this should use http.Server with Shutdown().

6. All repos - No request logging/middleware: There's no logging of incoming requests, which makes debugging harder.

7. All repos - No CORS headers: While the agent is the interface and CORS may not be strictly needed, it could be useful for browser-based agents.

8. All repos - No rate limiting: The spec mentions "plans, limits" but there's no rate limiting implemented.

9. All repos - No MCP connector: The spec says "Also speaks Model Context Protocol at /mcp" but none of the repos have this.

10. All repos - No audit logs: The spec says "audit logs built in from the start" but none of the repos have this.

11. All repos - OTP code generation has a bias: int(b[0])%1000000 - since b[0] is a byte (0-255), b[0]%1000000 is just b[0], so the OTP code is always between 000000 and 000255. This is a significant security bug! The OTP should use more random bytes to cover the full 000000-999999 range.

Wait, let me re-read the OTP generation:

b[0] is a single byte, range 0-255. int(b[0])%1000000 = int(b[0]) since 255 < 1000000. So the OTP is always between 000000 and 000255. This is a critical security bug! Only 256 possible codes instead of 1,000,000.

The fix should use all 4 bytes:

Wait, but int(b[0])<<24 could be negative if b[0] >= 128 on a 32-bit system... Actually in Go, int is at least 32 bits, and b[0] is a uint8. int(b[0]) is always 0-255, so int(b[0])<<24 is 0 to 255*16777216 = 4278190080, which fits in an int32 (max 2147483647)... no, 4278190080 > 2147483647, so on a 32-bit system this would overflow. But on a 64-bit system (which is the norm), int is 64 bits, so it's fine.

Actually, let me think more carefully. We should use uint32 to be safe:

This is the cleanest approach. But it requires importing encoding/binary.

Actually, an even simpler approach:

Or without binary:

This is a critical security bug that exists in ALL FOUR repos. Let me fix it in all of them.

12. All repos - rand.Read error not checked: In auth.go, rand.Read(b) errors are not checked in GenerateToken, GenerateOTP, and GenerateHandle. While crypto/rand.Read practically never fails, it's good practice to check the error.

13. All repos - No graceful shutdown: As mentioned above.

14. hookrelay - handleListEndpoints N+1 problem: Loading all deliveries for each endpoint just to count them is very inefficient. Should add a CountDeliveries method to the store.

Let me focus on the most impactful improvements:

1. Fix the OTP generation bug (critical security) - in all 4 repos
2. Add graceful shutdown - in all 4 repos
3. Fix the N+1 query in hookrelay - add a CountDeliveries method

Let me also check if there are any other issues I should address. Let me look at the auth files for the other repos to confirm the same OTP bug exists.

Now I have a clear picture. Let me plan the changes:

1. Fix OTP generation bug in all 4 repos - the GenerateOTP function only uses b[0] which gives only 256 possible codes. Fix to use all 4 bytes.

2. Add graceful shutdown to all 4 repos - replace http.ListenAndServe with http.Server + signal handling + Shutdown().

3. Fix N+1 query in hookrelay - add CountDeliveries method to store and use it in handleListEndpoints and handleGetEndpoint.

Let me also check the VERSION files for all repos to see if I should bump versions.

Now I have a complete picture. Let me plan the changes:

Changes to make across all 4 repos:

1. Fix OTP generation bug (critical security)

In all 4 repos, auth.go has:

b[0] is a single byte (0-255), so int(b[0])%1000000 = int(b[0]) which is always 0-255. This means only 256 possible OTP codes instead of 1,000,000.

Fix: Use all 4 bytes to generate the full range:

2. Add graceful shutdown to all 4 repos

Replace http.ListenAndServe with http.Server + signal handling + Shutdown().

3. Fix N+1 query in hookrelay

Add CountDeliveries method to store and use it in handleListEndpoints and handleGetEndpoint.

4. Bump versions

• hookrelay: 0.1.0 → 0.1.1
• taskpilot: 0.1.1 → 0.1.2
• notable: 0.1.1 → 0.1.2
• linksmith: 0.1.0 → 0.1.1

Let me start implementing these changes. I'll do all repos in parallel where possible.

Let me start with the auth.go fix for all 4 repos, then the main.go graceful shutdown for all 4 repos, then the hookrelay-specific N+1 fix.

For the auth.go fix, the change is the same in all repos - just the GenerateOTP function. Let me use replace_in_file for each.

For the main.go graceful shutdown, I need to rewrite the main function to use http.Server with signal handling. The pattern is the same across all repos, just with different service names and env var prefixes.

For the hookrelay N+1 fix, I need to:

1. Add CountDeliveries method to store.go
2. Update handleListEndpoints and handleGetEndpoint in api.go to use it

Let me start implementing.

All four repos have been successfully updated and pushed. Let me summarize what was done:

1. hookrelay (commit cb2d0dd, version 0.1.0 → 0.1.1):

   • Fixed OTP generation security bug
   • Added graceful shutdown
   • Fixed N+1 query (CountDeliveries method + updated handlers)

2. taskpilot (commit fdadf3d, version 0.1.1 → 0.1.2):

   • Fixed OTP generation security bug
   • Added graceful shutdown

3. notable (commit b60159e, version 0.1.1 → 0.1.2):

   • Fixed OTP generation security bug
   • Added graceful shutdown

4. linksmith (commit cce81e2, version 0.1.0 → 0.1.1):

   • Fixed OTP generation security bug
   • Added graceful shutdown

All repos pass go test ./..., go vet ./..., and CGO_ENABLED=0 go build.

The task is complete. Let me call _success.