Repository navigation
chore: fill or remove RSR template placeholders - #43
Conversation
openssf-compliance.yml fails when any of the thirteen files it checks
still contains a {{PLACEHOLDER}} token. This clears them.
Three kinds of change, no invention:
The "TEMPLATE INSTRUCTIONS (delete this block before publishing)" comment
is deleted. The template says to delete it, and it is where every legend
line lives -- so a large share of the reported tokens were the file
documenting its own placeholders, not real unfilled fields.
Tokens derivable from the repository are filled: owner and repo from the
git remote, project name, year, forge, main branch, contact email.
PGP and website lines are removed rather than filled, because nothing
true could go in them. https://github.com/<user>.gpg returns HTTP 200 for
every account; with no key uploaded the body is a stub reading "This user
hasnt uploaded any GPG keys". No key is published for either account
here, and commit signing in this estate is SSH, which is unrelated. Only
one repository in the estate has a domain, so {{WEBSITE}} likewise has no
correct value. The template sanctions this: "Optional: Remove sections
that dont apply (e.g. PGP if you dont use it)." A security policy telling
a researcher to encrypt to a key that does not exist is worse than one
that does not mention encryption.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Jonathan D.A. Jewell <6759885+hyperpolymath@users.noreply.github.com>
Four causes, none an unfilled field. STATE.a2ml prose that wraps mid-phrase, so a phrase-level replacement missed it. The bare token names the concept; it is now the word. An OpenSSF badge whose URL needs a project ID this repository does not have. The badge is removed rather than given an invented ID: a badge pointing at a nonexistent project is a broken image making a claim. CHANGELOG entries describing earlier placeholder work. Rewriting a changelog to say something it did not say would be wrong, so the sentence is kept and only the braces go -- the gate greps for the token shape and cannot tell a record of past work from present debt. Residual squisher-corpus identity that the previous pass did not reach. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Jonathan D.A. Jewell <6759885+hyperpolymath@users.noreply.github.com>
|
|
Note Automatic reviews are paused because your trial's included automatic processing has been used for this period. Upgrade now, or comment "Gitar review" to run a review anytime. CI failed: The CI build failed due to an unfilled placeholder token in .machine_readable/META.a2ml and a CodeQL configuration error attempting to analyze JavaScript/TypeScript in a non-JS repository.OverviewTwo distinct errors caused CI failures across jobs: an OpenSSF compliance check failure due to unfilled template placeholders in FailuresOpenSSF Compliance Placeholder Check (confidence: high)
CodeQL Analysis Configuration Error (confidence: high)
Summary
Code Review ✅ Approved 1 resolved / 1 findingsFills and removes RSR template placeholders to satisfy compliance checks, but leaves the README template note self-contradictory regarding token replacement instructions. ✅ 1 resolved✅ Quality: README template note became self-contradictory after substitution
Tip Comment OptionsDisplay: compact → Showing less information. Comment with these commands to change the behavior for this request:
Was this helpful? React with 👍 / 👎 | Gitar |
…ws (#65) Restore working CI through shared source repairs and immutable dependency locks. The standards reusable workflows now use e9fa12b965897f485f8ec143e048a424b52f689a from metadatastician/standards#764. The shared scanner fetches and verifies the exact revision used for its cache key, including cache hits. SSH mirrors verify independently approved host fingerprints before exporting strict SSH configuration or pushing. Optional mirror credentials are mapped explicitly. Canonical metadata is under .machine_readable/descriptiles/. Session coordination and custom checks are actual Nickel records. CI installs checksum-verified Nickel and evaluates them; envelope-aware typechecking has both passing and deliberately failing controls. The inactive E2E scaffold is a template, and the local E2E command reports that implementation is required. Action dependencies use GitHub's runtime-enforced actions.lock mechanism. The authoritative gh actions-lock tool verifies the exact commits, including transitive dependencies: https://github.com/github/gh-actions-lock#how-it-works . Inline mutable-looking tags do not imply mutable execution under this mechanism. Sonar's inline-SHA findings were classified individually after full lock verification; no rule-wide exclusion was added. Review disposition: the scanner/cache and mirror trust findings are fixed at source. Retired AGENTIC and anchor references and the directory-tree documentation are corrected. The earlier request to restore 8f2ee508 is superseded: that revision predates these security fixes. Local initialisation explicitly defers Nickel validation when Nickel is unavailable; CI always provisions it and requires validation. Also consumes the merged K9 archive repair (#43) and the merged CI/CD suite manifest gate (#5), whose valid/invalid controls pass. Pull-request triggers include stacked PR targets. This PR remains stacked on #64. The .envrc project name deliberately identifies this instantiated scientific repository. Restoring a template placeholder here would make the active project configuration incorrect; reusable minting behavior belongs in rsr-template-repo. Validation: workflow conventions and session Nickel evaluation pass; empty/malformed Nickel controls prove rejection. The CI/CD suite now runs real maintained A2ML/K9 structure validators, and the source tests establish their failure behavior. This is not a proof of full K9 semantics or production deployment. --------- Signed-off-by: Jonathan D.A. Jewell <6759885+hyperpolymath@users.noreply.github.com> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Co-authored-by: Jonathan D.A. Jewell <6759885+hyperpolymath@users.noreply.github.com> Co-authored-by: Mistral Vibe <vibe@mistral.ai> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: Claude Opus 5 <noreply@anthropic.com> Co-authored-by: rhodibot <rhodibot@hyperpolymath.dev> Co-authored-by: Arena Agent <bot@arena.ai>




openssf-compliance.ymlfails when any of the thirteen files it checks still contains a{{PLACEHOLDER}}token. This clears them, with no invention.TEMPLATE INSTRUCTIONS (delete this block before publishing)comment — the template says to delete it, and it is where every legend line lived, so a large share of the reported tokens were the file documenting its own placeholders.https://github.com/<user>.gpgreturns HTTP 200 for every account; with no key uploaded the body is a stub reading "This user hasn't uploaded any GPG keys". No key is published for either account, and commit signing here is SSH — unrelated. The template sanctions this: "Optional: Remove sections that don't apply (e.g. PGP if you don't use it)."\n\nA security policy telling a researcher to encrypt to a key that does not exist is worse than one that does not mention encryption.\n\nWhere applicable, this also fixes a misrouted advisory URL. A previousjust initcopied already-filled templates out ofsquisher-corpus, so 51 repositories directed vulnerability reports tohyperpolymath/squisher-corpus— a repository unrelated to the code being reported, meaning the affected maintainer never saw them. Repointed from this repository's own git remote.