agentic-devtools 0.2.142__tar.gz → 0.2.144__tar.gz

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/.devcontainer/devcontainer.json

@@ -36,7 +36,7 @@
 	// "forwardPorts": [],
 
 	// Use 'postCreateCommand' to run commands after the container is created.
-	"postCreateCommand": "pip install -e '.[dev]'",
+	"postCreateCommand": "pip install -e '.[dev]' && git config core.hooksPath .githooks",
 
 	// Configure tool-specific properties.
 	"features": {

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/.github/agents/agdt.address-copilot-review.evaluate-and-respond.agent.md

@@ -11,8 +11,8 @@ $ARGUMENTS
 You **MUST** consider the user input before proceeding (if not empty).
 
 The expected input is the body of the `@copilot` trigger comment posted on the PR, which
-contains a structured list of review comment links (and any suppressed comment excerpts),
-optional CI failure links, and a reference to this skill.
+contains collapsible `<details>` blocks with review comment links (and any suppressed
+comment excerpts), optional CI failure links, and a reference to this skill.
 
 ## Purpose
 
@@ -27,7 +27,8 @@ environment:
 4. After all comments are processed, post a **single structured summary comment** on the PR.
 5. If CI failures are listed, also address those (delegating to the CI repair sub-skill
    instructions).
-6. Commit, push, reply to threads, resolve threads, and re-request Copilot review.
+6. Commit and push. The post-push automation handles thread resolution, review re-request,
+   and squashing.
 
 **Scope limitations** (CI repair mode):
 
@@ -37,6 +38,26 @@ environment:
 - Does NOT execute PR-sourced scripts.
 - Does NOT approve or merge the PR.
 
+**Copilot Cloud Agent Restrictions**:
+
+- Cannot force-push (`git push --force`) — only regular `git push` is allowed.
+- Cannot resolve PR review comment threads via UI/API.
+- Cannot merge PRs.
+
+**Pre-push Hook Behavior**:
+
+Git hooks are pre-configured via `copilot-setup-steps.yml` — no manual setup needed.
+After running `git push`, the pre-push hook automatically runs targeted checks (ruff format,
+ruff check, per-file coverage, mypy, test structure validation). This can take up to
+**2 minutes**. Do NOT interrupt the push or re-push during this time. If the push is
+rejected because ruff reformatted files (look for "❌ Files were reformatted by ruff" in
+the output or `.pre-push-output.log`), stage the reformatted files, amend the commit, and
+push again.
+
+You do NOT need to manually run `ruff format`, `ruff check`, or `mypy` before pushing.
+The pre-push hook handles all of this automatically. Focus only on code changes that
+address the review feedback.
+
 ## Instructions
 
 Follow the detailed instructions in the corresponding prompt file

agentic_devtools-0.2.144/.github/copilot-setup-steps.yml

@@ -0,0 +1,3 @@
+steps:
+  - run: pip install 'agentic-devtools[dev]'
+  - run: git config core.hooksPath .githooks

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/.github/prompts/agdt.address-copilot-review.evaluate-and-respond.prompt.md

@@ -17,12 +17,42 @@
 >   changes — this is not "executing a PR-sourced script/binary".
 > - If `AGDT_CI_REPAIR_MODE=1` is set, these constraints are mandatory
 >   (defense-in-depth).
+>
+> **⚠️ COPILOT CLOUD AGENT RESTRICTIONS**:
+>
+> - Do NOT force-push (`git push --force` or `git push --force-with-lease`).
+> - Do NOT attempt to resolve PR review comment threads — the post-push automation handles this.
+> - Do NOT attempt to re-request Copilot review — the post-push automation handles this.
+> - Do NOT merge the PR.
 
 You are a senior software engineer addressing feedback from a GitHub Copilot pull request
 review. Follow this workflow systematically, completing each phase before proceeding.
 
 ---
 
+## Pre-push Hook & Linting
+
+> ⚠️ **Git hooks are pre-configured** via `copilot-setup-steps.yml`. You do NOT need to
+> set up hooks or run linting/formatting manually.
+>
+> After running `git push`, the pre-push hook automatically runs targeted checks
+> (ruff format, ruff check, per-file coverage, mypy, test structure validation).
+> This can take up to **2 minutes**. Do NOT interrupt the push or re-push during this time.
+> You can monitor progress by:
+>
+> - Checking `.pre-push-output.log` in the repo root
+> - Running status commands in a separate terminal
+>
+> If the push is rejected because ruff reformatted files (look for
+> "❌ Files were reformatted by ruff" in the output or `.pre-push-output.log`),
+> stage the reformatted files, amend the commit, and push again.
+>
+> You do NOT need to manually run `ruff format`, `ruff check`, or `mypy` before pushing.
+> The pre-push hook handles all of this automatically. Focus only on code changes that
+> address the review feedback.
+
+---
+
 ## Tooling Priority
 
 **Always prefer `agdt-*` commands** from agentic-devtools over raw alternatives:
@@ -30,16 +60,13 @@ review. Follow this workflow systematically, completing each phase before procee
 | Operation | Preferred (agdt-*) | Fallback (raw) |
 |-----------|-------------------|----------------|
 | Reply to review comments | `agdt-gh-reply-to-review-comments` | `gh api .../comments/{id}/replies` per comment |
-| Resolve review threads | `agdt-gh-resolve-review-threads` | GraphQL `resolveReviewThread` mutation per thread |
-| Request Copilot re-review | `agdt-gh-request-copilot-review` | `gh api .../requested_reviewers -X POST` |
-| Run targeted tests | `agdt-test-pattern`, `agdt-test-file` | Install `agentic-devtools` from PyPI (not the PR branch: `pip install agentic-devtools`), then use `agdt-test-pattern` |
+| Run targeted tests | `agdt-test-pattern`, `agdt-test-file` | Install `agentic-devtools[dev]` from PyPI (not the PR branch: `pip install 'agentic-devtools[dev]'`), then use `agdt-test-pattern` |
 | Stage changes | `agdt-git-stage` | `git add` |
-| Commit & push | `agdt-git-save-work` | Install `agentic-devtools` from PyPI (not the PR branch: `pip install agentic-devtools`), then use `agdt-git-save-work` |
-| Force push | `agdt-git-force-push` | `git push --force-with-lease` |
+| Commit & push | `agdt-git-save-work` | Install `agentic-devtools[dev]` from PyPI (not the PR branch: `pip install 'agentic-devtools[dev]'`), then use `agdt-git-save-work` |
 
 > **CI Repair Note**: This workflow requires `agentic-devtools` commands to be
 > available. If they are not installed, install from a trusted source —
-> **not** from the PR branch (`pip install agentic-devtools`). Do **not** fall
+> **not** from the PR branch (`pip install 'agentic-devtools[dev]'`). Do **not** fall
 > back to running `pytest` directly or using raw `git commit`/`git push`; both
 > violate project policy (see `.github/copilot-instructions.md`).
 
@@ -52,25 +79,31 @@ Extract identifiers from the `@copilot` trigger comment.
 ### Trigger Comment Format
 
 ```text
-@copilot
+@copilot - The Code Review Agent just left this [Review](https://github.com/{owner}/{repo}/pull/{pr_number}#pullrequestreview-{review_id}) that needs to be evaluated and addressed.
+
 <!-- copilot-trigger:{review_id} -->
 
-[Review](https://github.com/{owner}/{repo}/pull/{pr_number}#pullrequestreview-{review_id})
+<details>
+     <summary>Instructions</summary>
+
+You are an [`.github/agents/agdt.address-copilot-review.evaluate-and-respond.agent.md`](https://github.com/{owner}/{repo}/blob/main/.github/agents/agdt.address-copilot-review.evaluate-and-respond.agent.md) agent, your prompt can be found here: [`.github/prompts/agdt.address-copilot-review.evaluate-and-respond.prompt.md`](https://github.com/{owner}/{repo}/blob/main/.github/prompts/agdt.address-copilot-review.evaluate-and-respond.prompt.md).
 
-## Comments
+</details>
+
+<details>
+     <summary>Comments</summary>
 
 - [Comment #1 - filename.py (1)](https://github.com/{owner}/{repo}/pull/{pr_number}#pullreviewcomment-{id})
 - [Comment #2 - filename.py (2)](https://github.com/{owner}/{repo}/pull/{pr_number}#pullreviewcomment-{id})
 - Comment #3 - models.py (1): "body text" (suppressed comment)
+</details>
 
-## CI Failures
+<details>
+     <summary>CI Failures</summary>
 
 - ❌ [check-name](https://github.com/{owner}/{repo}/actions/runs/{run_id}/jobs/{job_id}) — `conclusion`
 - ❌ check-name-without-url — `conclusion`
-
-## Instructions
-
-Follow `.github/agents/agdt.address-copilot-review.evaluate-and-respond.agent.md`
+</details>
 ```
 
 ### Parse Out
@@ -83,7 +116,7 @@ Follow `.github/agents/agdt.address-copilot-review.evaluate-and-respond.agent.md
 | `review_id` | `<!-- copilot-trigger:... -->` marker or Review URL fragment | `4019856282` |
 | `visible_comments` | `[Comment #N - file (F)](url)` list items | List of `{nc, file, nf, url, comment_id}` |
 | `suppressed_comments` | `Comment #N - file (F): "body" (suppressed comment)` items | List of `{nc, file, nf, body}` |
-| `ci_failures` | `## CI Failures` section (if present) | List of `{name, url (optional), conclusion}` |
+| `ci_failures` | `<summary>CI Failures</summary>` section (if present) | List of `{name, url (optional), conclusion}` |
 
 Extract `comment_id` from each visible comment URL:
 `https://github.com/{owner}/{repo}/pull/{pr_number}#pullreviewcomment-{comment_id}`
@@ -92,8 +125,8 @@ Extract `comment_id` from each visible comment URL:
 
 - If the trigger comment is malformed or missing identifiers, check environment
   variables `REPAIR_PR_NUMBER`, `REPAIR_REVIEW_URL`, and `REPAIR_HEAD_SHA`.
-- If the `## CI Failures` section is absent, treat `ci_failures` as empty.
-- If `review_id` is present but there are no comment links under `## Comments`,
+- If the CI Failures `<details>` section is absent, treat `ci_failures` as empty.
+- If `review_id` is present but there are no comment links under Comments,
   fetch review comments directly:
   `gh api "repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments"` and
   continue using that list (`id`, `path`, `body`, `diff_hunk`).
@@ -113,7 +146,7 @@ gh api "repos/{owner}/{repo}/pulls/comments/{comment_id}" \
 
 | Field | Use |
 |-------|-----|
-| `id` | Database ID — needed for replying and resolving threads |
+| `id` | Database ID — needed for replying |
 | `path` | File path the comment is on |
 | `line` | Line number (may be `null` for file-level comments) |
 | `body` | Full comment text (the reviewer's feedback) |
@@ -185,7 +218,7 @@ Assign a confidence level for each suggestion:
 
 ## Phase 5: Handle CI Failures (if present)
 
-If the `## CI Failures` section is present in the trigger comment, address those failures
+If the CI Failures section is present in the trigger comment, address those failures
 before making code changes for review comments.
 
 The CI pipeline gates are:
@@ -292,6 +325,14 @@ agdt-git-save-work --skip-stage
 agdt-task-wait
 ```
 
+> ⚠️ After `git push`, the pre-push hook runs targeted checks automatically. This can
+> take up to **2 minutes**. Do NOT interrupt or re-push during this time. Check
+> `.pre-push-output.log` or use a separate terminal to monitor progress.
+>
+> If the push is rejected because ruff reformatted files (look for "❌ Files were
+> reformatted by ruff" in the output or `.pre-push-output.log`), stage the reformatted
+> files, amend the commit, and push again.
+
 Choose the commit **type** based on the nature of the changes:
 
 | Type | When to use |
@@ -309,18 +350,10 @@ COMMIT_FULL=$(git log -1 --format="%H")
 COMMIT_SHORT=$(git log -1 --format="%h")
 ```
 
-### Verify Push Was Successful
-
-```bash
-REMOTE_SHA=$(gh pr view {pr_number} --repo {owner}/{repo} --json headRefOid --jq '.headRefOid')
-```
-
-If `REMOTE_SHA` ≠ `COMMIT_FULL`, run `agdt-git-force-push` + `agdt-task-wait`, then re-verify.
-
 ### Edge Case: No Addressable Comments and No CI Failures
 
 If every comment was ❌ and there are no CI failures, skip the commit step and proceed
-directly to Phase 7 (replies).
+directly to Phase 7 (summary comment).
 
 ---
 
@@ -341,7 +374,7 @@ The comment MUST:
 6. Present CI failure resolutions in a separate table (if applicable)
 
 Include commit metadata and the commit link only when code changes were made.
-Include the CI section only when the trigger comment contained `## CI Failures`.
+Include the CI section only when the trigger comment contained CI Failures.
 
 ```markdown
 <!-- copilot-agent-result -->
@@ -388,7 +421,7 @@ Use exactly these decision labels (emoji + keyword) for programmatic parsing:
 
 ### CI Failures Table Rules
 
-- Only include the `### CI Failures Addressed` section if the trigger comment contained a `## CI Failures` section
+- Only include the `### CI Failures Addressed` section if the trigger comment contained a CI Failures section
 - Status values: `✅ Fixed` or `⚠️ Unable to fix` (with explanation)
 - If no CI failures existed, omit this section entirely
 
@@ -441,31 +474,13 @@ agdt-gh-reply-to-review-comments --pr {pr_number} --repo {owner}/{repo} \
 
 ---
 
-## Phase 9: Resolve Review Threads
-
-```bash
-agdt-gh-resolve-review-threads --pr {pr_number} --repo {owner}/{repo} \
-  --review-id {review_id}
-```
-
----
-
-## Phase 10: Re-request Copilot Review
-
-```bash
-agdt-gh-request-copilot-review --pr {pr_number} --repo {owner}/{repo}
-```
-
-If `verified` is `false`, retry once after 10 seconds.
-
----
-
 ## Error Handling
 
 - **Command failures**: If any `agdt-gh-*` command exits with a non-zero code, print
   stderr and handle as described in the relevant phase.
-- **Thread already resolved**: `agdt-gh-resolve-review-threads` handles this silently.
 - **Comment reply fails**: `agdt-gh-reply-to-review-comments` handles retries internally.
+- **Push rejected by hook**: If the push fails because ruff reformatted files (indicated
+  by "❌ Files were reformatted by ruff" in the hook output), stage the reformatted
+  files, amend the commit, and push again.
 - **No visible comments and no CI failures**: If the review has zero visible comments
-  and no CI failures, post the summary comment with suppressed-only rows if applicable,
-  then proceed to Phase 9 and 10.
+  and no CI failures, post the summary comment with suppressed-only rows if applicable.

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentic-devtools
-Version: 0.2.142
+Version: 0.2.144
 Summary: Agentic devtools integrate Jira, DevOps & more
 Author: ayaiayorg
 License-Expression: MIT

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/agentic_devtools/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.2.142'
-__version_tuple__ = version_tuple = (0, 2, 142)
+__version__ = version = '0.2.144'
+__version_tuple__ = version_tuple = (0, 2, 144)
 
 __commit_id__ = commit_id = None

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/agentic_devtools/cli/ci/github_provider.py

@@ -276,6 +276,9 @@ def _build_repair_comment(
     but don't begin with @copilot have been observed to trigger agent
     sessions unreliably.
 
+    Output uses collapsible ``<details>`` blocks to keep PR page real estate
+    compact while still providing full context on expansion.
+
     Args:
         head_sha: Current HEAD SHA for context.
         repair_type: ``"review"``, ``"ci"``, or ``"both"``.
@@ -289,24 +292,70 @@ def _build_repair_comment(
     Returns:
         Comment body string beginning with ``@copilot``.
     """
-    parts: list[str] = ["@copilot"]
-
     has_review = repair_type in ("review", "both")
     has_review_comments = has_review and bool(review_comments)
     has_ci = repair_type in ("ci", "both") and bool(failed_checks)
 
+    # --- First line: @copilot with inline summary ---
+    review_url = ""
+    if has_review and review_id and repository_full_name and "/" in repository_full_name and pr_number:
+        review_url = f"https://github.com/{repository_full_name}/pull/{pr_number}#pullrequestreview-{review_id}"
+
+    if has_review and review_url:
+        parts: list[str] = [
+            f"@copilot - The Code Review Agent just left this [Review]({review_url})"
+            " that needs to be evaluated and addressed."
+        ]
+    elif has_ci:
+        parts = ["@copilot - CI failures detected that need to be addressed."]
+    else:
+        parts = ["@copilot"]
+
+    # --- Dedup marker ---
     if has_review and review_id:
-        # Dedup marker so the agent can identify which review triggered this
+        parts.append("")
         parts.append(f"<!-- copilot-trigger:{review_id} -->")
 
-        if repository_full_name and "/" in repository_full_name and pr_number and review_id:
-            review_url = f"https://github.com/{repository_full_name}/pull/{pr_number}#pullrequestreview-{review_id}"
-            parts.append("")
-            parts.append(f"[Review]({review_url})")
+    # --- Fallback for empty context ---
+    has_review_context = has_review and bool(review_id)
+    if not review_comments and not failed_checks and not has_review_context:
+        parts.append("")
+        parts.append(f"Please review the PR and fix any issues found. Current HEAD: `{head_sha[:8]}`.")
+        parts.append("")
+        parts.append("---")
+        parts.append(f"*Automated repair dispatch for commit `{head_sha[:8]}` (type: {repair_type})*")
+        return "\n".join(parts)
+
+    # --- Instructions <details> block ---
+    if repair_type == "ci":
+        skill = "agdt.address-copilot-review.ci-repair.agent.md"
+        prompt_skill = "agdt.address-copilot-review.ci-repair.prompt.md"
+    else:
+        skill = "agdt.address-copilot-review.evaluate-and-respond.agent.md"
+        prompt_skill = "agdt.address-copilot-review.evaluate-and-respond.prompt.md"
+
+    if repository_full_name and "/" in repository_full_name:
+        agent_url = f"https://github.com/{repository_full_name}/blob/main/.github/agents/{skill}"
+        prompt_url = f"https://github.com/{repository_full_name}/blob/main/.github/prompts/{prompt_skill}"
+        agent_link = f"[`.github/agents/{skill}`]({agent_url})"
+        prompt_link = f"[`.github/prompts/{prompt_skill}`]({prompt_url})"
+    else:
+        agent_link = f"`.github/agents/{skill}`"
+        prompt_link = f"`.github/prompts/{prompt_skill}`"
+
+    parts.append("")
+    parts.append("<details>")
+    parts.append("     <summary>Instructions</summary>")
+    parts.append("")
+    parts.append(f"You are an {agent_link} agent, your prompt can be found here: {prompt_link}.")
+    parts.append("")
+    parts.append("</details>")
 
+    # --- Comments <details> block ---
     if has_review_comments:
         parts.append("")
-        parts.append("## Comments")
+        parts.append("<details>")
+        parts.append("     <summary>Comments</summary>")
         parts.append("")
 
         basename_paths: dict[str, set[str]] = {}
@@ -331,39 +380,21 @@ def _build_repair_comment(
                 else:
                     parts.append(f"- {label}")
 
+        parts.append("</details>")
+
+    # --- CI Failures <details> block ---
     if has_ci:
         parts.append("")
-        parts.append("## CI Failures")
+        parts.append("<details>")
+        parts.append("     <summary>CI Failures</summary>")
         parts.append("")
         for check in failed_checks:
-            # Only use html_url from the API response — the check run ID does not
-            # match the Actions workflow run ID, so omit the link entirely when
-            # html_url is absent to avoid emitting a broken /runs/{id} URL.
             job_url = check.html_url
             if job_url:
                 parts.append(f"- ❌ [{check.name}]({job_url}) — `{check.conclusion}`")
             else:
                 parts.append(f"- ❌ {check.name} — `{check.conclusion}`")
-
-    has_review_context = has_review and bool(review_id)
-    if not review_comments and not failed_checks and not has_review_context:
-        parts.append("")
-        parts.append(f"Please review the PR and fix any issues found. Current HEAD: `{head_sha[:8]}`.")
-        parts.append("")
-        parts.append("---")
-        parts.append(f"*Automated repair dispatch for commit `{head_sha[:8]}` (type: {repair_type})*")
-        return "\n".join(parts)
-
-    # Choose the appropriate skill based on what triggered the repair
-    if repair_type == "ci":
-        skill = "agdt.address-copilot-review.ci-repair.agent.md"
-    else:
-        skill = "agdt.address-copilot-review.evaluate-and-respond.agent.md"
-
-    parts.append("")
-    parts.append("## Instructions")
-    parts.append("")
-    parts.append(f"Follow `.github/agents/{skill}`")
+        parts.append("</details>")
 
     return "\n".join(parts)
 

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/agentic_devtools/cli/ci/pipeline/snapshot.py

@@ -253,7 +253,8 @@ def _count_unresolved_prior_threads(
     """Count unresolved Copilot review threads from commits before HEAD.
 
     Returns the count of review comments from prior-commit Copilot reviews
-    that would be eligible for resolution.
+    whose threads are actually unresolved per the GitHub API. If thread-state
+    data is unavailable, falls back to counting all non-synthetic comments.
     """
     prior_copilot_reviews = [
         r
@@ -266,15 +267,35 @@ def _count_unresolved_prior_threads(
     if not prior_copilot_reviews:
         return 0
 
+    # Fetch actual thread resolution states from GitHub GraphQL API
+    thread_statuses: dict[int, tuple[bool, bool]] | None = None
+    list_thread_states = getattr(provider, "list_review_thread_states", None)
+    if callable(list_thread_states):
+        try:
+            thread_statuses = list_thread_states(pr_number)
+        except Exception:
+            # Fail closed: if we can't determine resolution status, fall
+            # through to counting all comments (existing behavior).
+            thread_statuses = None
+
+    # Count only comments whose thread is NOT resolved (when status is known)
     total_unresolved = 0
     for prior_review in prior_copilot_reviews:
         try:
             comments = provider.list_review_comments(pr_number, prior_review.id)
-            # Exclude synthetic review-body entries — they have no real GitHub
-            # thread and must not inflate the unresolved count.
-            total_unresolved += sum(1 for c in comments if c.id >= 0)
+            for c in comments:
+                if c.id < 0:
+                    continue  # Skip synthetic review-body entries
+                if thread_statuses is not None:
+                    is_resolved, _has_reply = thread_statuses.get(c.id, (False, False))
+                    if not is_resolved:
+                        total_unresolved += 1
+                else:
+                    # Fallback: no thread status info, count all
+                    total_unresolved += 1
         except Exception:
-            return max(1, total_unresolved + 1)
+            # Fail closed for this review: count 1 unknown thread
+            total_unresolved += 1
     return total_unresolved
 
 

{agentic_devtools-0.2.142 → agentic_devtools-0.2.144}/tests/unit/cli/ci/github_provider/test__build_repair_comment.py

@@ -67,7 +67,8 @@ class TestBuildRepairComment:
             review_id=456,
         )
         assert "@copilot" in body
-        assert "## Comments" in body
+        assert "<details>" in body
+        assert "<summary>Comments</summary>" in body
         # Comment #1 - first comment on foo.py → foo.py (1)
         assert "[Comment #1 - foo.py (1)]" in body
         assert _COMMENT_1.html_url in body
@@ -79,7 +80,7 @@ class TestBuildRepairComment:
         assert _COMMENT_OTHER.html_url in body
 
     def test_review_repair_includes_review_link(self) -> None:
-        """Trigger comment includes a direct link to the parent review."""
+        """Trigger comment includes a direct link to the parent review in first line."""
         body = _build_repair_comment(
             head_sha="abc123def456",
             repair_type="review",
@@ -182,7 +183,7 @@ class TestBuildRepairComment:
         assert "Comment #3" in body  # next visible comment
 
     def test_ci_repair_lists_failed_checks(self) -> None:
-        """CI failures are listed with ❌ markers."""
+        """CI failures are listed with ❌ markers in a details block."""
         checks = [
             CheckRunStatus(id=1, name="Targeted Checks ✅", status="completed", conclusion="failure"),
             CheckRunStatus(id=2, name="Smart Module Tests ✅", status="completed", conclusion="failure"),
@@ -194,7 +195,7 @@ class TestBuildRepairComment:
             review_comments=[],
         )
         assert "@copilot" in body
-        assert "## CI Failures" in body
+        assert "<summary>CI Failures</summary>" in body
         assert "❌ Targeted Checks ✅" in body
         assert "❌ Smart Module Tests ✅" in body
 
@@ -250,7 +251,7 @@ class TestBuildRepairComment:
         assert "`failure`" in body
 
     def test_both_repair_includes_review_and_ci_sections(self) -> None:
-        """Combined repair includes both ## Comments and ## CI Failures sections."""
+        """Combined repair includes both Comments and CI Failures details blocks."""
         checks = [CheckRunStatus(id=1, name="tests", status="completed", conclusion="failure")]
         body = _build_repair_comment(
             head_sha="abc123def456",
@@ -258,8 +259,8 @@ class TestBuildRepairComment:
             failed_checks=checks,
             review_comments=[_COMMENT_1],
         )
-        assert "## Comments" in body
-        assert "## CI Failures" in body
+        assert "<summary>Comments</summary>" in body
+        assert "<summary>CI Failures</summary>" in body
 
     def test_empty_context_includes_fallback(self) -> None:
         """When no comments and no failures, a fallback message is shown with SHA."""
@@ -285,7 +286,7 @@ class TestBuildRepairComment:
         )
         assert "<!-- copilot-trigger:456 -->" in body
         assert "[Review](https://github.com/owner/repo/pull/42#pullrequestreview-456)" in body
-        assert "## Instructions" in body
+        assert "<summary>Instructions</summary>" in body
         assert "agdt.address-copilot-review.evaluate-and-respond.agent.md" in body
 
     def test_fallback_does_not_include_skill_reference(self) -> None:
@@ -296,7 +297,7 @@ class TestBuildRepairComment:
             failed_checks=[],
             review_comments=[],
         )
-        assert "## Instructions" not in body
+        assert "<summary>Instructions</summary>" not in body
 
     def test_review_only_references_evaluate_and_respond_skill(self) -> None:
         """Review-only repair references the evaluate-and-respond skill."""
@@ -334,15 +335,15 @@ class TestBuildRepairComment:
         assert "agdt.address-copilot-review.ci-repair.agent.md" not in body
 
     def test_instructions_section_present_when_context_exists(self) -> None:
-        """## Instructions section appears when there are comments or CI failures."""
+        """Instructions details block appears when there are comments or CI failures."""
         body = _build_repair_comment(
             head_sha="abc123def456",
             repair_type="review",
             failed_checks=[],
             review_comments=[_COMMENT_1],
         )
-        assert "## Instructions" in body
-        assert "Follow `.github/agents/" in body
+        assert "<summary>Instructions</summary>" in body
+        assert "`.github/agents/" in body
 
     def test_per_file_counter_resets_for_each_file(self) -> None:
         """nF counter resets to 1 for each new file path."""
@@ -370,7 +371,7 @@ class TestBuildRepairComment:
         assert "Comment #2 - pkg_b/__init__.py (1)" in body
 
     def test_no_old_section_headers_in_new_format(self) -> None:
-        """Old-style headers (## Copilot Review Feedback, ## CI Failure Context) are gone."""
+        """Old-style ## headers are replaced by <details> blocks."""
         checks = [CheckRunStatus(id=1, name="lint", status="completed", conclusion="failure")]
         body_ci = _build_repair_comment(
             head_sha="abc123def456",
@@ -387,3 +388,44 @@ class TestBuildRepairComment:
         assert "## Copilot Review Feedback" not in body_review
         assert "## CI Failure Context" not in body_ci
         assert "### Comment" not in body_review
+        assert "## Comments" not in body_review
+        assert "## CI Failures" not in body_ci
+        assert "## Instructions" not in body_review
+
+    def test_ci_only_first_line_mentions_ci_failures(self) -> None:
+        """CI-only repair starts with @copilot mentioning CI failures."""
+        checks = [CheckRunStatus(id=1, name="lint", status="completed", conclusion="failure")]
+        body = _build_repair_comment(
+            head_sha="abc123def456",
+            repair_type="ci",
+            failed_checks=checks,
+            review_comments=[],
+        )
+        assert body.startswith("@copilot - CI failures detected")
+
+    def test_review_first_line_mentions_review(self) -> None:
+        """Review repair starts with @copilot mentioning the review."""
+        body = _build_repair_comment(
+            head_sha="abc123def456",
+            repair_type="review",
+            failed_checks=[],
+            review_comments=[_COMMENT_1],
+            repository_full_name="owner/repo",
+            pr_number=42,
+            review_id=456,
+        )
+        assert body.startswith("@copilot - The Code Review Agent just left this [Review]")
+
+    def test_instructions_contains_links_when_repo_provided(self) -> None:
+        """Instructions block contains full URL links when repository_full_name is provided."""
+        body = _build_repair_comment(
+            head_sha="abc123def456",
+            repair_type="review",
+            failed_checks=[],
+            review_comments=[_COMMENT_1],
+            repository_full_name="owner/repo",
+            pr_number=42,
+            review_id=456,
+        )
+        assert "https://github.com/owner/repo/blob/main/.github/agents/" in body
+        assert "https://github.com/owner/repo/blob/main/.github/prompts/" in body