CLAUDE.md — Claude Code persistent instructions

Read

@ai/CLAUDE.md

for project intelligence, owner decisions, architecture vision, grading tiers, TRON dashboard specs, and everything from the spec review sessions.

Read and follow

@ai/AGENTS.md

for project philosophy, architecture, code style, commit conventions, and the full agent contract.

Read

@ai/DEV.md

for development workflow, TDD discipline, and definition of done.

Quick reference

make install                  # Install all dependencies (dev + dashboard + superset)
pre-commit install            # Install git hooks (required before first commit)
uv run pytest                 # Run Python unit tests
pre-commit run --all-files    # Run all pre-commit checks
make code-check               # Lint (ruff) + typecheck (mypy)
make robot-dryrun             # Validate Robot tests without execution

Rules

Always run tests after changes.
```
uv run pytest
```
for Python,
```
make robot-dryrun
```
for Robot Framework.
Always run
pre-commit run --all-files
before committing. Never bypass hooks.
Don't remove functionality without explicit approval.
Be verbose in CLI output — when running commands, show what's happening.
TDD is mandatory. Write a failing test first, then implement, then refactor.
Atomic commits only. One idea per commit, using conventional format:
```
<type>: <summary>
```
.
Never bundle unrelated changes in a single commit.
Never mix formatting changes with logic changes.
Type hints are required on all new Python code. mypy must pass.
Use
RETURN
(not
[Return]
) in Robot Framework keywords.

Architecture guardrails

```
src/rfc/
```
is the single source of truth for all Python code.
```
robot/
```
is the single home for all Robot Framework test suites.
Never duplicate logic outside these directories.
Listeners are always active in make targets — don't strip them.

Environment

Copy

.env.example

.env

and edit before running integration tests. Key variables:

OLLAMA_ENDPOINT

DEFAULT_MODEL

DATABASE_URL

. See

ai/DEV.md

§ Environment Configuration for the full list.

Quick reference

make install                  # Install all dependencies (dev + dashboard + superset)
pre-commit install            # Install git hooks (required before first commit)
uv run pytest                 # Run Python unit tests
pre-commit run --all-files    # Run all pre-commit checks
make code-check               # Lint (ruff) + typecheck (mypy)
make robot-dryrun             # Validate Robot tests without execution

Rules

Always run tests after changes.

uv run pytest

for Python,

make robot-dryrun

for Robot Framework.

Always run
pre-commit run --all-files
before committing. Never bypass hooks.

Don't remove functionality without explicit approval.

Be verbose in CLI output — when running commands, show what's happening.

TDD is mandatory. Write a failing test first, then implement, then refactor.

Atomic commits only. One idea per commit, using conventional format:

<type>: <summary>

Never bundle unrelated changes in a single commit.

Never mix formatting changes with logic changes.

Type hints are required on all new Python code. mypy must pass.

Use
RETURN
(not
[Return]
) in Robot Framework keywords.

CLAUDE.md — Claude Code persistent instructions

CLAUDE.md — Claude Code persistent instructions

Quick reference

Rules

Architecture guardrails

Environment

Related Skills

Markdown Converter

Nano Banana Pro

1password

CLAUDE.md — Claude Code persistent instructions

Quick reference

Rules

Architecture guardrails

Environment