Test agent by reading README
Takeaway: I didn't really give it anything to work with here, just 'test' with no real task, so nothing came of it.
634 builds · this week 621
Feed
Explain how LLMs represent chain-of-thought reasoning
Takeaway: This was just a quick informational question, not a coding task, so there's not much to correct here.
8s0interventions0 cmt
Review Gemini CLI thinking-tag fix PR (pi-mono)
Takeaway: I'd start by pointing the agent straight at both the PR and the linked issue URLs together so it doesn't have to chain gh calls to discover the issue reference itself.
Read README file (test session)
Takeaway: I didn't actually give the agent a real task here, just sent 'test' twice, so there's nothing to evaluate.
10s0interventions0 cmt
Fix thinking-block code blocks displaying wrong colors in web UI
Takeaway: Next time I'd check upstream for a recent fix before diving into diagnosis and pushing a branch — it turned out badlogic/pi-mono had already patched this exact issue.
Brainstorm architecture for an agentic todo app
Takeaway: I'd probably scope the request more next time, since 'best ever' invited a broad conceptual answer rather than concrete code.
Explore building a todo app in pi-mono monorepo
Takeaway: I got sidetracked asking the agent about its own thinking blocks and context files instead of pushing forward on the actual todo app design, so next time I'd hold off on meta-questions until after getting a concrete plan or prototype out of it.
Explore pi-mono repo structure
Takeaway: This was just a test prompt, so there's not much to judge yet.
18s0interventions0 cmt
Investigate markdown list renumbering behavior in TUI/web-ui
Takeaway: I'd start by asking directly whether this is standard Markdown behavior before diving into a code investigation, since it turned out to be expected CommonMark semantics all along.
Investigate how pi passes context to Codex vs Anthropic
Takeaway: Next time I'd start by asking directly for the diff between the Anthropic and Codex provider adapters in pi-ai instead of grepping blindly through node_modules.
Test Ralph loop mechanics with a dummy task
Takeaway: I just wanted to see the Ralph loop mechanics in action, so a quick 'start a dummy loop' prompt was enough.
Test Ralph loop iteration workflow
Takeaway: I'd start by saying I just want to test one iteration manually rather than letting the loop auto-chain, since the tool seems to want to keep calling ralph_done on its own.
Debug alt+enter not working in Pi TUI under Warp terminal
Takeaway: Next time I'd ask directly for a side-by-side comparison of how Warp's raw escape sequences map onto both tools' key-matching code, since that's what actually mattered — Pi parses raw bytes while Codex/crossterm gets structured key+modifier events.
Locate provider-specific context file path
Takeaway: I just wanted a quick confirmation of which context file was in use, and the agent gave me the exact path immediately without extra fluff.
3s0interventions0 cmt
List loaded context files in pi agent
Takeaway: I just wanted a quick check on which context files were active, and the agent answered directly with the three paths.
Tune difficulty and polish a /ping terminal game extension (TypeScript)
Takeaway: Next time I'd start by just playing the game myself before asking for a review, since most of the real bugs (unwinnable speed, glitchy edge bounces, AI difficulty swings) only showed up once I actually tried to beat Pi.
Generate placeholder text sample
Takeaway: Next time I'd give the agent a clear topic, tone, and length upfront instead of just saying 'generate some text'.
5s0interventions0 cmt
Fix editor clearing draft text on queued steering messages in pi's TUI
Takeaway: Next time I'd ask upfront for both the fix and a git-history deep dive on the offending line, since that back-and-forth took longer than the fix itself.
Investigate and respond to GitHub issue about Excel add-in context handling
Takeaway: Next time I'd ask the agent up front to separate 'what's already documented' from 'what's an actual gap' so I get that split without a second pass.
Review and merge dependabot npm dependency bump PR
Takeaway: Next time I'd tell it upfront that auto-merge isn't enabled on this repo and branch protection requires an up-to-date branch, so it could go straight to triggering a dependabot rebase instead of hitting three failed merge attempts first.