feat: post success-path step comments via trusted credentials for change/test/checkup/e2e_fix

architecture.json

@@ -80,6 +80,21 @@
             "signature": "(text: Optional[str]) -> Optional[str]",
             "returns": "Optional[str]"
           },
+          {
+            "name": "extract_step_report",
+            "signature": "(text: Optional[str]) -> Optional[str]",
+            "returns": "Optional[str]"
+          },
+          {
+            "name": "normalize_step_comments_state",
+            "signature": "(raw: Any) -> Set[int]",
+            "returns": "Set[int]"
+          },
+          {
+            "name": "post_step_comment_once",
+            "signature": "(*, repo_owner: str, repo_name: str, issue_number: int, step_num: int, body: str, posted_steps: Set[int], cwd: Path) -> bool",
+            "returns": "bool"
+          },
           {
             "name": "_sanitize_comment_body",
             "signature": "(body: Optional[str], max_chars: int = 25_000) -> str",

pdd/agentic_change_orchestrator.py

@@ -29,6 +29,9 @@
     post_step_comment,
     set_agentic_progress,
     clear_agentic_progress,
+    extract_step_report,
+    normalize_step_comments_state,
+    post_step_comment_once,
 )
 from pdd.load_prompt_template import load_prompt_template
 from pdd.sync_order import (
@@ -1417,7 +1420,10 @@ def run_agentic_change_orchestrator(
         model_used = "unknown"
         github_comment_id = None
         worktree_path = None
-
+
+    step_comments_set: Set[int] = normalize_step_comments_state(state.get("step_comments"))
+    state["step_comments"] = sorted(step_comments_set)
+
     pddrc_context = _load_pddrc_context(cwd)
 
     context = {
@@ -1694,6 +1700,7 @@ def run_agentic_change_orchestrator(
                         output=step_output, cwd=cwd,
                     )
                     state["step_outputs"][str(step_num)] = f"FAILED: {step_output}"
+                    state["step_comments"] = sorted(step_comments_set)
                     save_workflow_state(cwd, issue_number, "change", state, state_dir, repo_owner, repo_name, use_github_state, github_comment_id)
                     return False, f"Aborting: {consecutive_provider_failures} consecutive steps failed — agent providers unavailable", total_cost, model_used, []
             else:
@@ -1717,6 +1724,7 @@ def run_agentic_change_orchestrator(
                     refreshed = _fetch_issue_updated_at(repo_owner, repo_name, issue_number)
                     if refreshed:
                         state["issue_updated_at"] = refreshed
+                state["step_comments"] = sorted(step_comments_set)
                 save_workflow_state(cwd, issue_number, "change", state, state_dir, repo_owner, repo_name, use_github_state, github_comment_id)
                 return False, f"Stopped at step {step_num}: {stop_reason}", total_cost, model_used, changed_files
             console.print(f"[yellow]Warning: Step {step_num} reported failure but continuing...[/yellow]")
@@ -1739,6 +1747,7 @@ def run_agentic_change_orchestrator(
                 refreshed = _fetch_issue_updated_at(repo_owner, repo_name, issue_number)
                 if refreshed:
                     state["issue_updated_at"] = refreshed
+            state["step_comments"] = sorted(step_comments_set)
             save_workflow_state(cwd, issue_number, "change", state, state_dir, repo_owner, repo_name, use_github_state, github_comment_id)
             return False, f"Stopped at step {step_num}: {stop_reason}", total_cost, model_used, changed_files
 
@@ -1771,6 +1780,7 @@ def run_agentic_change_orchestrator(
                 # Issue #467: Mark as FAILED instead of using step_num - 1
                 state["step_outputs"][str(step_num)] = f"FAILED: {step_output}"
                 # Don't advance last_completed_step — keep it at its current value
+                state["step_comments"] = sorted(step_comments_set)
                 save_workflow_state(cwd, issue_number, "change", state, state_dir, repo_owner, repo_name, use_github_state, github_comment_id)
                 return False, "Stopped at step 9: Implementation produced no file changes", total_cost, model_used, []
 
@@ -1925,6 +1935,28 @@ def run_agentic_change_orchestrator(
             consecutive_provider_failures = 0
             state["step_outputs"][str(step_num)] = step_output
             state["last_completed_step"] = step_num
+            try:
+                report_body = extract_step_report(step_output)
+                if not report_body:
+                    report_body = (
+                        f"_Step {step_num} completed; no `<step_report>` block "
+                        "returned by agent. Raw output retained in workflow state._"
+                    )
+                comment_body = (
+                    f"## Step {step_num}/13: {description}\n\n{report_body}"
+                )
+                post_step_comment_once(
+                    repo_owner=repo_owner,
+                    repo_name=repo_name,
+                    issue_number=issue_number,
+                    step_num=step_num,
+                    body=comment_body,
+                    posted_steps=step_comments_set,
+                    cwd=current_work_dir,
+                )
+            except Exception as exc:  # pylint: disable=broad-except
+                console.print(f"[yellow]post_step_comment_once failed: {exc}[/yellow]")
+            state["step_comments"] = sorted(step_comments_set)
         else:
             state["step_outputs"][str(step_num)] = f"FAILED: {step_output}"
 
@@ -1970,6 +2002,33 @@ def run_agentic_change_orchestrator(
                 instruction=s11_prompt, cwd=current_work_dir, verbose=verbose, quiet=quiet, timeout=timeout11, label=f"step11_iter{review_iteration}", max_retries=DEFAULT_MAX_RETRIES, reasoning_time=reasoning_time,
             )
             total_cost += s11_cost; model_used = s11_model; state["total_cost"] = total_cost
+            # Round-2 of Greg's review: gate the trusted Step 11 post on
+            # s11_success. A failed task (e.g. provider exhaustion) still
+            # returns step output, and posting a "completed" fallback would
+            # both mislead the user and mark this iteration's composite key
+            # (review_iteration*100 + 11) as already-posted in
+            # state["step_comments"]. On a later resume/retry the real
+            # successful Step 11 report would be silently deduped away.
+            if s11_success:
+                try:
+                    s11_report = extract_step_report(s11_output)
+                    if not s11_report:
+                        s11_report = (
+                            "_Step 11 completed; no `<step_report>` block returned "
+                            "by agent. Raw output retained in workflow state._"
+                        )
+                    post_step_comment_once(
+                        repo_owner=repo_owner,
+                        repo_name=repo_name,
+                        issue_number=issue_number,
+                        step_num=review_iteration * 100 + 11,
+                        body=f"## Step 11/13: Review (iteration {review_iteration})\n\n{s11_report}",
+                        posted_steps=step_comments_set,
+                        cwd=current_work_dir,
+                    )
+                    state["step_comments"] = sorted(step_comments_set)
+                except Exception as exc:  # pylint: disable=broad-except
+                    console.print(f"[yellow]post_step_comment_once failed: {exc}[/yellow]")
             if _review_loop_no_issues(s11_output):
                 if not quiet: console.print("   -> No issues found. Proceeding to PR.")
                 context["step11_output"] = s11_output; break
@@ -1989,6 +2048,30 @@ def run_agentic_change_orchestrator(
             total_cost += s12_cost; model_used = s12_model; state["total_cost"] = total_cost
             previous_fixes += f"\n\nIteration {review_iteration}:\n{s12_output}"
             state["previous_fixes"] = previous_fixes
+            # Round-2 of Greg's review: gate the trusted Step 12 post on
+            # s12_success. Same reasoning as Step 11 above — a failed fix
+            # task must not burn the composite key, otherwise a later
+            # successful retry of the same iteration would be deduped.
+            if s12_success:
+                try:
+                    s12_report = extract_step_report(s12_output)
+                    if not s12_report:
+                        s12_report = (
+                            "_Step 12 completed; no `<step_report>` block returned "
+                            "by agent. Raw output retained in workflow state._"
+                        )
+                    post_step_comment_once(
+                        repo_owner=repo_owner,
+                        repo_name=repo_name,
+                        issue_number=issue_number,
+                        step_num=review_iteration * 100 + 12,
+                        body=f"## Step 12/13: Fix (iteration {review_iteration})\n\n{s12_report}",
+                        posted_steps=step_comments_set,
+                        cwd=current_work_dir,
+                    )
+                    state["step_comments"] = sorted(step_comments_set)
+                except Exception as exc:  # pylint: disable=broad-except
+                    console.print(f"[yellow]post_step_comment_once failed: {exc}[/yellow]")
             save_result = save_workflow_state(cwd, issue_number, "change", state, state_dir, repo_owner, repo_name, use_github_state, github_comment_id)
             if save_result: github_comment_id = save_result; state["github_comment_id"] = github_comment_id
         if review_iteration >= MAX_REVIEW_ITERATIONS:
@@ -2097,8 +2180,28 @@ def run_agentic_change_orchestrator(
         if not s13_success:
              post_step_comment(repo_owner, repo_name, issue_number, 13, 13, "Create PR and link to issue", s13_output, cwd)
              console.print("[red]Step 13 (PR Creation) failed.[/red]")
+             state["step_comments"] = sorted(step_comments_set)
              save_workflow_state(cwd, issue_number, "change", state, state_dir, repo_owner, repo_name, use_github_state, github_comment_id)
              return False, "PR Creation failed", total_cost, model_used, changed_files
+        try:
+            s13_report = extract_step_report(s13_output)
+            if not s13_report:
+                s13_report = (
+                    "_Step 13 completed; no `<step_report>` block returned "
+                    "by agent. Raw output retained in workflow state._"
+                )
+            post_step_comment_once(
+                repo_owner=repo_owner,
+                repo_name=repo_name,
+                issue_number=issue_number,
+                step_num=13,
+                body=f"## Step 13/13: Create PR and link to issue\n\n{s13_report}",
+                posted_steps=step_comments_set,
+                cwd=current_work_dir,
+            )
+            state["step_comments"] = sorted(step_comments_set)
+        except Exception as exc:  # pylint: disable=broad-except
+            console.print(f"[yellow]post_step_comment_once failed: {exc}[/yellow]")
         pr_url = "Unknown"; url_match = re.search(r"https://github.com/\S+/pull/\d+", s13_output)
         if url_match: pr_url = url_match.group(0)
         if not quiet:

pdd/agentic_checkup_orchestrator.py

@@ -18,17 +18,20 @@
 import shutil
 import subprocess
 from pathlib import Path
-from typing import Dict, List, Optional, Tuple, Union
+from typing import Dict, List, Optional, Set, Tuple, Union
 
 from rich.console import Console
 
 from .agentic_common import (
     DEFAULT_MAX_RETRIES,
     _sanitize_comment_body,
     clear_workflow_state,
+    extract_step_report,
     load_workflow_state,
+    normalize_step_comments_state,
     post_pr_comment,
     post_step_comment,
+    post_step_comment_once,
     run_agentic_task,
     save_workflow_state,
     substitute_template_variables,
@@ -1029,6 +1032,11 @@ def run_agentic_checkup_orchestrator(
         if changed_files:
             context["files_to_stage"] = ", ".join(changed_files)
 
+    if state is None:
+        step_comments_set: Set[int] = set()
+    else:
+        step_comments_set = normalize_step_comments_state(state.get("step_comments"))
+
     # Step definitions (step 6 split into 6.1/6.2/6.3 sub-steps).
     steps: List[Tuple[Union[int, float], str, str]] = [
         (1,   "discover",         "Discovering project structure and tech stack"),
@@ -1069,6 +1077,7 @@ def _save_state() -> None:
             pr_owner=pr_owner,
             pr_repo=pr_repo,
             pr_head_sha=current_pr_head_sha if pr_mode else None,
+            step_comments=sorted(step_comments_set),
         )
         github_comment_id = save_workflow_state(
             cwd=cwd, issue_number=issue_number, workflow_type="checkup",
@@ -1078,6 +1087,43 @@ def _save_state() -> None:
             github_comment_id=github_comment_id,
         )
 
+    def _step_comment_key(step_num: Union[int, float], iteration: int = 1) -> int:
+        """Project (step_num, iteration) -> deterministic non-negative int.
+
+        Encoding ``iteration * 10000 + int(round(step_num * 10))`` handles
+        fractional steps (6.1 -> 61) and the iterated fix-verify loop.
+        """
+        return iteration * 10000 + int(round(float(step_num) * 10))
+
+    def _maybe_post_step_comment(
+        step_num: Union[int, float],
+        description: str,
+        step_output: str,
+        iteration: int = 1,
+    ) -> None:
+        """Post a trusted per-step success comment; log-and-continue on error."""
+        try:
+            report_body = extract_step_report(step_output)
+            if not report_body:
+                report_body = (
+                    f"_Step {step_num} completed; no `<step_report>` block "
+                    "returned by agent. Raw output retained in workflow state._"
+                )
+            comment_body = (
+                f"## Step {step_num}/{TOTAL_STEPS}: {description}\n\n{report_body}"
+            )
+            post_step_comment_once(
+                repo_owner=repo_owner,
+                repo_name=repo_name,
+                issue_number=issue_number,
+                step_num=_step_comment_key(step_num, iteration),
+                body=comment_body,
+                posted_steps=step_comments_set,
+                cwd=current_cwd,
+            )
+        except Exception as exc:  # pylint: disable=broad-except
+            console.print(f"[yellow]post_step_comment_once failed: {exc}[/yellow]")
+
     def _is_provider_failure(output: str) -> bool:
         return "All agent providers failed" in output
 
@@ -1087,6 +1133,8 @@ def _handle_step_result(
         output: str,
         cost: float,
         model: str,
+        description: str = "",
+        iteration: int = 1,
     ) -> Optional[Tuple[bool, str, float, str]]:
         """Process a step result — update context, save state.
 
@@ -1123,6 +1171,8 @@ def _handle_step_result(
             step_outputs[step_key] = output
             last_completed_step_to_save = step_num
             consecutive_provider_failures = 0
+            if description:
+                _maybe_post_step_comment(step_num, description, output, iteration)
         else:
             step_outputs[step_key] = f"FAILED: {output}"
             if _is_provider_failure(output):
@@ -1352,7 +1402,7 @@ def _post_pr_mode_final_report(final_step7_output: str) -> str:
 
         success, output, cost, model = result
 
-        abort = _handle_step_result(step_num, success, output, cost, model)
+        abort = _handle_step_result(step_num, success, output, cost, model, description=description)
         if abort is not None:
             return abort
 
@@ -1395,7 +1445,7 @@ def _post_pr_mode_final_report(final_step7_output: str) -> str:
                 return (False, f"Missing prompt template: {template_name}", total_cost, last_model_used)
 
             success, output, cost, model = result
-            abort = _handle_step_result(step_num, success, output, cost, model)
+            abort = _handle_step_result(step_num, success, output, cost, model, description=description)
             if abort is not None:
                 return abort
             if not success and _is_provider_failure(output):
@@ -1444,7 +1494,7 @@ def _post_pr_mode_final_report(final_step7_output: str) -> str:
 
             success, output, cost, model = result
             nofix_step7_output = output
-            abort = _handle_step_result(7, success, output, cost, model)
+            abort = _handle_step_result(7, success, output, cost, model, description=desc7)
             if abort is not None:
                 return abort
         else:
@@ -1585,7 +1635,10 @@ def _post_pr_mode_final_report(final_step7_output: str) -> str:
                     return (False, f"Missing prompt template: {tmpl_name}", total_cost, last_model_used)
 
                 success, output, cost, model = result
-                abort = _handle_step_result(step_num, success, output, cost, model)
+                abort = _handle_step_result(
+                    step_num, success, output, cost, model,
+                    description=description, iteration=fix_verify_iteration,
+                )
                 if abort is not None:
                     return abort
 
@@ -1854,7 +1907,7 @@ def _post_pr_mode_final_report(final_step7_output: str) -> str:
                 return (False, f"Missing prompt template: {template_name}", total_cost, last_model_used)
 
             success, output, cost, model = result
-            abort = _handle_step_result(8, success, output, cost, model)
+            abort = _handle_step_result(8, success, output, cost, model, description=desc8)
             if abort is not None:
                 return abort
 
@@ -1903,6 +1956,7 @@ def _build_state(
     pr_owner: Optional[str] = None,
     pr_repo: Optional[str] = None,
     pr_head_sha: Optional[str] = None,
+    step_comments: Optional[List[int]] = None,
 ) -> Dict:
     """Build a serialisable state dict for persistence.
 
@@ -1938,4 +1992,5 @@ def _build_state(
         "pr_owner": pr_owner,
         "pr_repo": pr_repo,
         "pr_head_sha": pr_head_sha,
+        "step_comments": list(step_comments) if step_comments else [],
     }