Turn Pi into el Gentleman: a senior-architect development harness with SDD/OpenSpec, subagents, strict TDD evidence, review guardrails, and skill discovery.
gentle-pi has 9 tools with mixed quality. Naming is generally verb-based (read, bash, grep, find, ls, edit, write, codegraph, ask_user_choice), but several tools lack proper input schema documentation. The codegraph tool has a well-defined schema with constraints (operation enum, query minLength/maxLength, limit min/max). However, 5 of 9 tools (read, bash, grep, find, ls, edit, write) show minimal schema detail in the provided source, properties are present but lack type constraints, minLength, maxLength, or enum values. Descriptions are present but brief (10-50 chars for most tools). The bash tool is particularly concerning: it accepts arbitrary shell commands with no validation, sanitization, or safety constraints. Error handling is absent, no recovery guidance, no categorization of errors as retryable vs fatal. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (IRREVERSIBLE for bash, WRITE for edit/write). Output schemas are not documented.
Ask one single-select question with two to four ordered options. A custom response is available only when allowCustomResponse is explicitly enabled.
Execute bash commands in the workspace
Initialize, search, or explore the CodeGraph index for the current Pi workspace only. This tool never accepts a project path or shell command.
Edit file contents
Find files matching a pattern
Search for patterns in files
List directory contents
bash tool accepts arbitrary shell commands with no input validation, sanitization, or safety constraints. No description of what it does, when to use it, or consequences. Marked IRREVERSIBLE but no confirmation/dry-run pattern.
5 of 9 tools (read, bash, grep, find, ls, edit, write) lack proper input schema constraints. Properties exist but have no type, minLength, maxLength, enum, or pattern validation. LLMs cannot infer valid input ranges.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. bash is IRREVERSIBLE, edit/write are WRITE operations. Agents cannot reason about safety without annotations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Read file contents from the workspace
Write content to a file
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. Codegraph, read, bash, grep, find, ls, edit, write all lack return type documentation.
No error handling guidance. Tools do not categorize errors as retryable, user-fixable, or fatal. No recovery suggestions (e.g., 'file not found, try ls() to list available files'). Agents cannot self-correct.