refactor: standardize bot identity to shadow (match deequ) (#278)

* Standardize bot identity to shadow (match deequ)

Rename the bot's user-facing identity from deequ-bot to shadow so all three
migrated repos (deequ, python-deequ, dqdl) are uniform. The migration PR kept
the legacy deequ-bot name to avoid touching auto-approve.yml; this aligns it
with the shared engine's home project.

bot.name, the auto-approve clean marker, the workflows[] trigger entry, and
the caller workflow name all move in lockstep — changing any one alone breaks
the marker match and silently disables auto-approve.

- .shadow.yml: bot.name deequ-bot -> shadow
- auto-approve.yml: CLEAN_MARKER + workflows[] entry -> shadow / Shadow
- issue-bot.yml: workflow name PyDeequ Bot -> Shadow

(prompt_sm_prefix stays pydeequ-bot — that is the Secrets Manager namespace, not the bot identity.)

* Doc hygiene: restore caller comment + clarify model-pin override

- issue-bot.yml: re-add deequ's comment explaining BEDROCK_REPORTER/CRITIC_MODEL_ID
  are sourced from repo variables (empty falls back to engine defaults)
- .shadow.yml: note the BEDROCK_MODEL_ID secret overrides models.investigator
  (live opus-4-8), so the yaml value is the fallback; critic has no secret so it
  runs the yaml value

* Harden auto-approve: never self-approve guardrail changes

Add a check that skips auto-approval when a PR modifies .github/** or
.shadow.yml. Without it, a PR that weakens the approval gate (or repoints the
bot at a different engine) is approved by the very gate it edits. Paginates
listFiles so a large PR cannot slip a guarded file past a 30-item page.

Author: sudsali

.github/workflows/auto-approve.yml

@@ -2,7 +2,7 @@ name: Auto-Approve Clean PRs
 
 on:
   workflow_run:
-    workflows: [".github/workflows/base.yml", "PyDeequ Bot"]
+    workflows: [".github/workflows/base.yml", "Shadow"]
     types: [completed]
 
 permissions:
@@ -55,6 +55,21 @@ jobs:
               return;
             }
 
+            // Condition 0: never auto-approve a PR that changes the bot's own
+            // guardrails (workflows or engine config) — that lets a PR weaken
+            // the approval gate and be approved by the very gate it edits.
+            // Require a human for these.
+            const files = await github.paginate(github.rest.pulls.listFiles, {
+              owner, repo, pull_number: prNumber, per_page: 100
+            });
+            const guarded = files.find(f =>
+              f.filename.startsWith('.github/') || f.filename === '.shadow.yml'
+            );
+            if (guarded) {
+              core.info(`PR touches guarded path ${guarded.filename} — requires human review, skipping`);
+              return;
+            }
+
             // Condition 1: CI must have passed for this SHA
             const {data: workflowRuns} = await github.rest.actions.listWorkflowRunsForRepo({
               owner, repo, head_sha: sha, status: 'completed'
@@ -72,7 +87,7 @@ jobs:
               owner, repo, pull_number: prNumber
             });
 
-            const CLEAN_MARKER = '<!-- deequ-bot:clean -->';
+            const CLEAN_MARKER = '<!-- shadow:clean -->';
 
             const latestBot = reviews
               .filter(r => r.user.login === 'github-actions[bot]')

.github/workflows/issue-bot.yml

@@ -1,4 +1,4 @@
-name: PyDeequ Bot   # load-bearing: auto-approve.yml keys on this exact name
+name: Shadow   # load-bearing: auto-approve.yml keys on this exact name
 # To upgrade the engine, bump the SHA in BOTH `uses:` and `shadow_ref` below
 # (GitHub forbids expressions in `uses:`, so they can't share a variable).
 
@@ -54,6 +54,7 @@ jobs:
       KB_S3_BUCKET: ${{ secrets.KB_S3_BUCKET }}
       KB_S3_KEY: ${{ secrets.KB_S3_KEY }}
       BEDROCK_MODEL_ID: ${{ secrets.BEDROCK_MODEL_ID }}
+      # Repo variables (not secrets); empty falls back to engine defaults.
       BEDROCK_REPORTER_MODEL_ID: ${{ vars.BEDROCK_REPORTER_MODEL_ID }}
       BEDROCK_CRITIC_MODEL_ID: ${{ vars.BEDROCK_CRITIC_MODEL_ID }}
       SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}

.shadow.yml

@@ -7,7 +7,7 @@ codebase:
   language: python
 
 bot:
-  name: deequ-bot   # not pydeequ-bot: must render the <!-- deequ-bot:clean --> marker auto-approve.yml greps for
+  name: shadow   # renders the <!-- shadow:clean --> marker auto-approve.yml greps for
   attribution: "Reviewed by Shadow · github.com/sudsali/shadow"
   escalate_label: needs-human
   max_replies: 2