Skip to content

Point the agent docs at opt-solve and opt-trade, and record three traps - #64

Merged
raorjun merged 1 commit into
mainfrom
docs/optsim-agent-docs
Sep 24, 2026
Merged

raorjun merged 1 commit into
mainfrom
docs/optsim-agent-docs

Conversation

@raorjun

@raorjun raorjun commented Sep 21, 2026 •

Copy link
Copy Markdown
Member

What

This PR updates BobSim's agent docs for the OptSim work in #59, #61 and #63. Stacked on #63. Docs only.

The routing table in AGENTS.md and docs/README.md still described OptSim as sweeps plus a reverse lookup. So an agent asked to size a bar or compare two options would go to the wrong tool. They now name all three tools (opt-standard, opt-solve, opt-trade) and say which question each one answers.

  • docs/workflows.md lists the two new targets and where they cache.
  • docs/architecture.md names the two modules that the solver and the trade study share.

This PR adds three entries under Common agent mistakes. Each one cost real time during work on those PRs.

  • An OpenModelica -override can be accepted and do nothing. OpenModelica evaluates toe, camber and every mass and CG value at compile time. The parameter still reports isValueChangeable="true", and the runner gives no warning. Only the paths in RUNTIME_SAFE_PATHS are proven to follow an override.
  • OptSim is three tools. None of them finds a "best" car. A consumer must use VariantStore. It must not rewrite the sweep's committed _doe_config.yaml, which records the scope that opt-search needs.
  • Git Bash on Windows rewrites container paths. /workspace/... becomes C:/Program Files/Git/workspace/..., and the run stops in seconds. MSYS_NO_PATHCONV=1 fixes it.

How to check

Read the four files. This PR has no code. python -m pytest tests -q gives the same result as on #63, because no test checks doc contents. The claims are the ones that #61 and #63 proved by experiment, and the docs link to them.

Notes

The change is small on purpose. AGENTS.md has its own rule that a stale doc is worse than no doc. This PR adds no new doc file, so neither docs table needed a new row. The full explanation stays in docs/doe-reverse-engineering.md. These entries are pointers and traps.

Merge after #63. The OptSim page on the public site is updated separately in BobDyn/BobDocs.

@raorjun
raorjun added this pull request to stack #65 September 21, 2026 03:25
@raorjun
raorjun marked this pull request as ready for review September 21, 2026 03:31
@raorjun
raorjun force-pushed the docs/optsim-agent-docs branch from 410dcd3 to 5f39c91 Compare September 21, 2026 03:47
@raorjun
raorjun force-pushed the docs/optsim-agent-docs branch 3 times, most recently from 223cc83 to 3f86708 Compare September 24, 2026 01:30
Base automatically changed from feat/optsim-trade-study to main September 24, 2026 01:32
The routing table and the docs index still described OptSim as sweeps
plus a reverse lookup. So an agent asked to size a bar or compare two
options would go to the wrong tool. They now name all three tools
and say which question each one answers.

Three traps cost real time during this work. They are now under common
mistakes:

- OpenModelica accepts an override of toe, camber or any mass value,
  and the override silently does nothing.
- A consumer must not rewrite the sweep's committed _doe_config.yaml.
  It records the scope that opt-search needs.
- Git Bash on Windows rewrites the /workspace paths that make passes to
  Docker.
@raorjun
raorjun force-pushed the docs/optsim-agent-docs branch from 3f86708 to 51987bd Compare September 24, 2026 01:32
@raorjun
raorjun merged commit 4da577a into main Sep 24, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant