feat(cli): serve guidance from the CLI - #1382
Conversation
✨ Aspect Workflows Tasks📅 Sun Aug 16 19:31:04 UTC 2026 ❌ 2 failed tasks
|
83a3616 to
e1405cf
Compare
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
cbf302d to
a3476cb
Compare
a3476cb to
ff26d07
Compare
What changed
Adds
aspect prompts, a bundled prompt library for coding agents working with repositories.aspect prompts listlists the available guidance.aspect prompts show <PROMPT>prints one prompt as Markdown.aspect prompts detectinspects repository markers, recommends an ordered prompt sequence, and explains each recommendation;--jsonprovides the same information structurally.bazel run, detection usesBUILD_WORKSPACE_DIRECTORY, so it inspects the caller's checkout rather than the runfiles tree.The catalogue deliberately focuses on high-signal guidance: Rust, Go, Python, JavaScript/TypeScript, Scala, protobuf, container, rules_rs, rules_py, rules_oci, and Aspect Remote scenarios. Java and generic in-place Go/JavaScript upgrade prompts were removed. Shared foundations now cover only version precedence, root labels, Git ignore policy, and meaningful validation.
Prompt bodies are compiled Markdown assets. Repository AXL/config cannot redefine the catalogue or its policy.
aspect describeremains the resolved task and feature surface; it does not own prompt content.Why
Prompts need to be available before a repository has Bazel or Aspect configuration. Keeping scenario inference in the CLI makes them discoverable without allowing it to silently choose an agent's final artefact.
Validation
ASPECT_WRAPPER_SKIP=1 bazel test //crates/aspect-cli:testASPECT_WRAPPER_SKIP=1 bazel build //crates/aspect-cliLLMs used
Best from my tests: Claude with Opus 5, with high effort. It's also faster.
Codex,
gpt-5.6-terrawith medium effort was less capable. But, better I got better results with sol and high effortAt least for rust, Codex with terra seems to draw the wrong conclusions more often.