The cleanup command removes entire data directories under analysis/, but per-file analysis artifacts are intentionally stored alongside MCP data as {id}.analysis.md (see analysis/README.md:461). This will delete those analysis markdown files and can also wipe tracked historical analysis/**/data/ content, causing large unintended deletions in the PR. Instead, scope cleanup to the current run’s directory and delete only raw data file types (e.g., JSON) while preserving *.analysis.md and other markdown artifacts.

This cleanup deletes all analysis/**/data directories, which removes per-file analysis outputs (*.analysis.md) that are stored alongside MCP data by design (analysis/methodologies/ai-driven-analysis-guide.md:422). It can also delete previously committed historical data directories, producing a huge PR diff. Please limit deletion to the current date/type directory and remove only raw MCP payload files (while keeping *.analysis.md/markdown).

Removing entire analysis/**/data directories will also remove per-file analysis markdown ({id}.analysis.md) that is supposed to live next to each downloaded MCP file (analysis/README.md:461). It may also delete tracked historical data and create unintended deletions in the PR. Recommend deleting only raw data file extensions inside the current run’s data directory, preserving *.analysis.md and other markdown artifacts.

The find ... -name "data" -exec rm -rf cleanup removes per-file analysis artifacts that are stored in the data/ tree as *.analysis.md (analysis/templates/per-file-political-intelligence.md:24). It can also wipe historical analysis/<date>/data already committed to the repo. Please adjust cleanup to preserve markdown analysis files and avoid deleting outside the current run’s directory.

This command deletes all analysis/**/data directories, but the repo’s analysis protocol stores per-file analysis markdown alongside raw MCP data in data/ (analysis/README.md:435-462). That means this cleanup will delete required analysis artifacts and may also remove already-committed historical data. Please only delete raw data files (not *.analysis.md) and scope to the current run’s path.

This cleanup deletes every analysis/**/data directory, but the analysis methodology stores per-file analysis markdown in that data/ tree as {id}.analysis.md (analysis/methodologies/ai-driven-analysis-guide.md:422). It will therefore delete required analysis artifacts and may wipe tracked historical data. Please adjust to remove only raw MCP payload files (e.g., JSON) within the current run’s directory, preserving *.analysis.md and other markdown.

The cleanup deletes analysis/**/data directories, but per-file analysis artifacts are saved alongside the raw data as {id}.analysis.md under data/ (analysis/README.md:461). This will remove required analysis markdown and can also delete committed historical data directories, creating unintended PR deletions. Please scope cleanup to the current date/type and delete only raw data file types while preserving *.analysis.md/markdown artifacts.

This cleanup removes analysis/**/data directories, but translation analysis artifacts may need to reference per-file analyses stored inside data/ as {id}.analysis.md (analysis/templates/per-file-political-intelligence.md:24). Deleting the whole directory will also delete those markdown artifacts and can wipe tracked historical analysis/<date>/data content. Please delete only raw payload files (e.g., JSON) and scope to the current run’s directory, preserving markdown.

Rule 5 says raw MCP data/ subdirectories may be cleaned, but elsewhere in this guide per-file analysis is defined to live inside data/ as {id}.analysis.md (see “Conflict Resolution” table). As written, “clean data/ subdirectories” can be interpreted as deleting the entire directory and would remove required analysis artifacts. Please clarify that cleanup must only remove raw payload files (e.g., *.json) while preserving *.analysis.md and other markdown artifacts, or adjust the prescribed storage layout so analysis is not under data/.

This rule allows cleaning raw MCP data/ files, but the analysis protocol stores per-file analysis markdown alongside the MCP data inside data/ (analysis/README.md “Save analysis alongside data file”). Please tighten the wording so “cleaning data” cannot be interpreted as deleting the whole data/ directory, and explicitly require preserving *.analysis.md/markdown artifacts.

The cleanup command deletes all non-Markdown files under analysis/**/data/**, which will also remove already-committed historical MCP payloads (e.g., analysis/2026-03-28/data/meps/person-*.json). That would create very large diffs and break existing analysis references. Scope cleanup to the current workflow’s directory (e.g., analysis/${TODAY}/week-in-review/data/) and/or delete only newly-downloaded/untracked payloads (so tracked historical files are not removed).

The find analysis/ ... -delete cleanup will delete all non-Markdown files in every analysis/**/data/** folder, including historically committed MCP payloads. This risks huge PRs with mass deletions and leaves existing analysis referencing missing data files. Please scope deletion to analysis/${TODAY}/week-ahead/data/ (and/or only remove untracked/new downloads) instead of the whole analysis/ tree.

This cleanup deletes all non-Markdown files under analysis/**/data/** across the repo, including already-committed historical payloads. That can produce massive deletion diffs and break existing analysis files that reference those payload paths. Please scope deletion to the current run directory (e.g., analysis/${TODAY}/month-in-review/data/) or delete only newly-downloaded/untracked payloads.

The workflow now mandates analysis-only PRs, but also says to use safeoutputs___noop when MCP is completely unavailable and zero data is collected. That contradicts Rule 5’s “no workflow run wasted” requirement: even on MCP failure you can still commit an analysis artifact documenting the outage (and any troubleshooting context) via an analysis-only PR. Consider removing this exception or narrowing it to cases where PR creation itself is impossible.

The raw payload cleanup is scoped to the entire analysis/ tree, so it will delete non-Markdown files from historically committed analysis/**/data/** directories (not only today’s breaking run). This can generate massive deletion diffs and break existing analysis references. Please scope cleanup to analysis/${TODAY}/breaking/data/ and/or delete only newly-downloaded/untracked files.

This cleanup command operates on analysis/ globally and will delete non-Markdown files under all analysis/**/data/** directories, including payload files already committed in the repo for previous dates. That can create very large PR diffs and break existing analysis that references those files. Please scope deletion to analysis/${TODAY}/${ARTICLE_TYPE_SLUG}/data/ (and/or only remove untracked/new downloads).

In this cleanup block, ${TODAY} is used to build RUN_ANALYSIS_DIR before TODAY is (re)set later in the same step. If $TODAY is unset or differs from the intended UTC date, cleanup may target the wrong directory (or analysis//breaking). Compute/export TODAY once before this block (prefer UTC) and avoid reassigning it after the cleanup so the directory, branch name, and PR title all align.

Step 4c adds mandatory translation analysis artifacts, but earlier steps still safeoutputs___noop and exit 0 when (a) no English articles exist for the date or (b) translations already exist. Those early exits happen before this step runs, so the workflow can still produce no committed analysis, contradicting Rule 5 and the PR description's “noop runs produce an analysis artifact”. Move baseline analysis creation earlier (before any noop/exit), and replace these noop paths with an analysis-only PR when any analysis/artifacts (even the baseline) are required.

In the "no English articles found" branch, the workflow calls safeoutputs___create_pull_request but does not exit 0 (or otherwise stop) afterwards. Execution will continue with empty $ARTICLE_TYPES/$NEEDS_TRANSLATION, likely causing incorrect behavior and potentially attempting additional PR creation. Add an explicit exit/return immediately after creating the analysis-only PR.

In the "all translations already exist" branch, the workflow calls safeoutputs___create_pull_request but does not exit 0 afterwards. The remaining steps will still run even though the analysis-only PR is already created, which can lead to errors or duplicate work. Add an explicit exit/return immediately after creating the analysis-only PR.

Same as the earlier noop branch: safeoutputs___create_pull_request({ ... }) is shown inside a bash block, which is likely to be executed as shell code and fail. Move the tool call into a javascript block (or otherwise ensure the instructions can be executed without mixing shell and tool-call syntax).