Repository navigation
Conversation
The docs said scan makes one source edit, in an HTML shell, only after a successful post. Since scan started writing the production marker all three clauses are wrong: there are two edits, the marker goes into a JSX root, and it is written before the post so an offline build still gets the flag. Agents read these files and treat an undisclosed capability as misrepresentation, so leaving the old text in place is worse than the gap it described. Also documents the marker in the command-reference rows for `scan` and `setup`, in the first-scan summary, and in the manual-placement step for code roots — which is the step that shipped a widget tag with no marker beside it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
Documentation clearly describes both the disclosure widget update and new production marker integration. 🎯 Quality: 93% Elite · 📦 Size: Small 📈 This month: Your 63rd PR — above team average · Averaging Excellent |
daniloradovic
approved these changes
Aug 24, 2026
4 tasks
Contributor
Author
|
Superseded by #171 (same disclosure, rebased onto current main after the credential-file doc updates). |
Contributor
Author
|
Closing in favor of #171. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up to #165. That PR gave
scana new source edit and did not update the docs that describe whatscanwrites, whichCLAUDE.mdtreats as the expensive kind of inaccuracy: "agents audit the shipped docs, so inaccuracies inAGENT-INSTALL.mdcost trust and cause refusals."Three separate claims went stale, all in one sentence:
|| truebuild still gets the flagCorrected there and in the
scan/setuprows of the command-reference table, the first-scan summary, and the manual-placement step for code roots — which is precisely the step that told an agent to add a widget tag with no marker beside it.README.mdcarried the same claims in its command summary and The disclosure widget section; both updated.Not changed
The install prompt. It already reads "authorize its widget and production build hooks" and "leave dev builds unchanged", which the
import.meta.env.PRODgate satisfies exactly — the marker is inert in dev. No new authorization is needed, so the tested artifact stays untouched.Separate finding, not fixed here
CLAUDE.mdstates the prompt "appears in three places that must stay identical:README.md,GETTING-STARTED.md, andfield-test/prompt.txt". They have drifted:GETTING-STARTED.mdandfield-test/prompt.txtmatch each other.README.md:11is a different prompt — hosted-workspace variant, addsPATCHSTACK_ENVIRONMENT=sandbox, authorizes "runtime protection source edits", and says regular dependency where the other two say dev dependency.Two consequences worth a decision: the README ships a prompt the field-test harness never exercises, and the tested prompt asks for a devDependency install, which contradicts README's own warning that "deployments that prune dev dependencies need the package in
dependencies" — the shape a real Lovable project has.Reconciling them means editing the adversarial surface, so it belongs in its own change with a field-test run behind it, not bundled into a docs correction.
Before merging
Docs-only, no code.
npm test1315 passed / 6 skipped,npm run typecheckclean. This touchesAGENT-INSTALL.md, so perCLAUDE.mdit wantsnode field-test/run.mjs --persona hostile --rounds 3— not run, same as #165.🤖 Generated with Claude Code