Smoke test Pi agent extension behaviors (read/write/subagent)
Session from pine-of-glass-sessions on Hugging Face by thomasmustier · MIT · condensed by Coders Talk
Click a segment to jump to that moment
3msession
0interventions
by library · 2d
GOAL
Verify a Pi coding agent extension's basic capabilities: reading files, writing/deleting temp files, invoking subagents, and reasoning in a scratchpad before acting.
- +0Prompt · +0Read a few files, then write a file to /tmp/ and delete it - extension test
- +0mAgent did · +0mRead AGENTS.md (missing), package.json, and README.md, then created and deleted a temp file under /tmp/ successfully
- +1mPrompt · +1mThink for a bit, write a long poem about the sea to a file, delete it, then invoke a subagent to do something trivial
- +1mAgent did · +1mWrote a 91-line sea poem to a temp file, deleted it, listed available subagents, then delegated a trivial smoke-test task to a subagent which confirmed success
- +3mPrompt · +3mThink for a bit in your scratchpad, then read some files
- +3mAgent did · +3mRead package.json and README.md continuations plus docs; one path (docs/context-budget.md) didn't exist, so it listed docs/ and read testing.md, pi-contextimate.md, and the Codex context-accounting note instead
- +3mOutcome · +3mExtension test completed: read/write/delete, scratchpad reasoning, and subagent delegation all verified working across three prompts
VERDICT
I'd start by just naming exactly which extension behaviors I want tested up front (read, write/delete, subagent, scratchpad) rather than spreading it across three prompts. The agent handled the missing-file error on its own without needing correction, so nothing here really needed my intervention — it self-recovered by listing the docs directory when a path didn't exist.