flow-context
This project uses flow for automation. When asked to build, test, run, deploy, lint, generate, or perform any dev task — or to run any one-off shell command — prefer the flow MCP tools over raw Bash s
SKILL.md.claude/skills/flow-context/SKILL.md
GitHubThis repository uses flow for all development automation. The mcp__flow__* MCP tools are available — prefer them over raw Bash for anything runnable, so it executes with the workspace's environment and secrets and is captured in flow's execution history.
Running work — pick the closest tool
- Named task? (build, test, lint, validate, generate, deploy, …) →
mcp__flow__list_executablesto find it, thenmcp__flow__executewith its verb + name. Don't hand-roll a shell command a flow executable already covers. - Arbitrary one-off command? (a
git ...,go test ./..., a script) →mcp__flow__run_commandwith the command and a shortlabel. This is preferred over rawBash: it runs with workspace env/secrets and lands inflow logswith provenance. Passcommands(array) +mode(serial/parallel) to run several in one call. - One-off is Python? →
mcp__flow__run_pythonwithcodeand a shortlabel, rather thanpython -cor a scratch.pyfile. flow resolves the workspace's virtualenv, so imports see the project's dependencies, and tracebacks report real line numbers. - Something richer than one command? (a serial/parallel batch, an HTTP
request) →mcp__flow__run_executablewith an inlinespec. - Only fall back to
Bashwhen a command genuinely shouldn't be recorded or flow isn't the right tool (e.g. interactive/TTY programs).
Key executables in this repo
Common refs: test unit, test e2e, lint, validate, generate, build binary. Use mcp__flow__list_executables to discover current names — don't assume.
Context & authoring
- Call
mcp__flow__get_infoat the start of a session, or when you need schema URLs to author.flowfiles. - Author or edit flow files with
mcp__flow__write_flowfile(validates against the schema server-side) rather than writing YAML by hand. - Runs are scoped to the current workspace by default;
run_command/run_python/run_executableaccept an optionalworkspaceto target another without switching the global current workspace. - To review what you've run this session, call
mcp__flow__get_execution_logswithmine: true;source/session/statusfilter history more broadly (alsoflow logs --source mcp --session <id>).