Conversation
…ifest detection finds nothing (fixes #20) Manifest detection is evidence-only by design: no package.json script, no pyproject/ruff/pytest config, no go.mod or Cargo.toml means every *_cmd comes back blank. That part is correct and stays. What was wrong is what happened next — the script signed off with "edit .flow/config.md to fill in your project commands", the skill said the user supplies what the script cannot infer, and the template told them to add one by hand. In a script-style Python repo (a requirements.txt, a venv, a runnable test_*.py) that is homework handed to the user for work the agent can do in one step: the evidence is right there on disk. - scripts/init.js: extract doneMessage(), which reports which *_cmd fields the manifests did not evidence and addresses closing that gap to the agent — infer from the repo, verify the command runs, write it into config.md. A re-run over an existing config detected nothing and so asks for nothing. detectStackCommands is untouched; it still never invents an unevidenced command. - commands/init.md: new Step 4 — glob for the evidence a manifest did not carry (venv interpreter, test files, dev entrypoint, Makefile/justfile target), run each inferred command before writing it, then Edit it into config.md. Report and health move to Steps 5 and 6. - references/config-template.md, SETUP.md, commands/health.md, README.md: a blank field now means "nothing in the repo supported one", not "your turn". - scripts/init.test.js: captureStdout() plus coverage for each doneMessage branch and the issue's own repro end-to-end. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Member
Author
|
Test run of the Nova→dsh→Claude pipeline. Closing to redo with live-watch delegation; same fix will be resubmitted. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes flykit-cc/flow#3
What changed
When
/flow:init's manifest detection comes up empty (a script-style Python repo withrequirements.txt+ venv but nopyproject.toml/package.json), the init script and skill used to hand the user homework: "edit.flow/config.mdto fill in your project commands" / "add one by hand if the detection missed it". This PR makes the agent close that gap instead — infer the commands from the repo, verify each one actually runs, and write the verified value into.flow/config.md.scripts/init.js— extracteddoneMessage(); the closing output now reports which*_cmdfields the manifests did not evidence and addresses the follow-up to the agent ("inspect the repo… infer each blank *_cmd, verify it actually runs, and write the verified value into .flow/config.md"), never to the user. A re-run over an existing config asks for nothing.detectStackCommandsis untouched — the evidence-only invariant ("never invents a command that isn't evidenced by a file on disk") is preserved and still pinned by tests.commands/init.md— new Step 4: glob for the evidence a manifest didn't carry (venv interpreter,test_*.py/tests/, dev entrypoint,Makefile/justfile/Taskfile.ymltarget), run each inferred command before writing it (never leave a dev server running), thenEditthe verified value into.flow/config.md. AddedEdit, Globtoallowed-tools.references/config-template.md,SETUP.md,commands/health.md,README.md— a blank*_cmdnow means "nothing in the repo supported one", not "your turn".scripts/init.test.js— 7 new tests: eachdoneMessage()branch, and the issue's own repro end-to-end (requirements.txt + venv +test_*.py⇒ output directs the agent, never the user).Version
flow bumped 0.6.0 → 0.7.0 (minor: new behavior) in
plugins/flow/.claude-plugin/plugin.json,plugins/flow/package.json, and.claude-plugin/marketplace.json.Testing
cd plugins/flow && npm test— 180/180 pass (was 173; +7 new)node --check plugins/flow/scripts/init.jsandinit.test.js— pass./scripts/check-plugin-versions.sh --since main— pass (ok flow: bumped 0.6.0 -> 0.7.0)init.jsagainst the issue's repro repo — output carries the agent-directed inference step, no "edit config.md by hand" text