claude-dev-env 8.15.0 → 8.15.1

package/docs/codex-compatibility.md

@@ -36,39 +36,10 @@ The default `.claude` target uses `CODEX_HOME`. A named profile or another
 target uses `<managed-root>/.codex`, so launch that profile with the same
 `CODEX_HOME` value.
 
-## Per-message context injection (manual wiring)
-
-Codex reads the same `hookSpecificOutput.additionalContext` shape as Claude,
-on its own `UserPromptSubmit` event. `hooks/session/style_reminder_prompt.py`
-works for both, as is.
-
-The install already points `$CODEX_HOME/hooks` at the shared hooks folder.
-Once installed, the script sits at
-`$CODEX_HOME/hooks/session/style_reminder_prompt.py`.
-
-Add this by hand to `$CODEX_HOME/hooks.json` (merge it into an existing file):
-
-```json
-{
-  "hooks": {
-    "UserPromptSubmit": [
-      {
-        "command": "python3 $CODEX_HOME/hooks/session/style_reminder_prompt.py"
-      }
-    ]
-  }
-}
-```
-
-The package installer does not register this optional context hook. Add it by
-hand when the host should receive that context.
-
-## Luna fast-mode guard
-
-Commit `0f21faf8` retired `luna_fast_mode_gate.py` with the other blocking
-policy hooks. The package ships no replacement, so a Codex `hooks.json` needs
-no entry for it, and the spawn tier a Luna spawn asks for is the caller's to
-hold.
+## Luna spawn tier
+
+No hook gates the tier a Luna spawn asks for, so a Codex `hooks.json` needs no
+entry for it, and the caller holds that tier.
 
 ## Roots and safety
 

package/docs/references/team-advisor-skill.md

@@ -9,7 +9,6 @@
 | `.agents/skills/team-advisor/SKILL.md` | Sole-consumer bind wiring and constraints |
 | `advisor-tool.md` | Consult timing, hard rule, how to treat advice |
 | `~/.claude/_shared/advisor/advisor-protocol.md` | Session identity, host bind, floor, lifecycle |
-| `agents/session-advisor.md` | ENDORSE / CORRECTION / PLAN / STOP |
 
 ## When to use
 

package/docs/worker-completion-gate.md

@@ -30,4 +30,3 @@ Mark `completed` only when every row passes.
 ## Relationship to other rules
 
 - `long-horizon-autonomy` covers acting on what you have and not ending a turn on a promise. This gate names the specific completion condition: workers finished and their results merged.
-- `skills/pr-converge/reference/state-schema.md` defines the run-state records a worker's result lands in before the task closes.

package/hooks/advisory/migration_safety_advisor.py

@@ -52,8 +52,9 @@ def main() -> None:
         operations = ", ".join(found_unsafe)
         advisory_message = (
             f"MIGRATION SAFETY: Contains {operations}. "
-            "Post-launch, model changes MUST be backwards-compatible. "
-            "Verify this won't break running instances during deployment."
+            "After launch, model changes stay backwards-compatible, because running "
+            "instances keep reading the old schema until the deployment finishes. "
+            "Check that this migration keeps those instances working."
         )
         advisory_payload = {
             "systemMessage": advisory_message,

package/hooks/advisory/refactor_guard.py

@@ -264,9 +264,9 @@ def build_refactor_advisory_context(refactor_description: str, file_path: str) -
     """Build guidance that names the Edit stage and changed-surface rule."""
     return (
         f"[HOOK ADVISORY] Refactor guard — {refactor_description} in {file_path}. "
-        "Edit-stage guidance: Only modify lines already changed in the current git diff. "
-        "Ask the user for explicit approval first. If the user approves, "
-        "create the bypass token then retry."
+        "Edit-stage guidance: This edit went through, and it changes lines outside the "
+        "current git diff. Keep further edits to lines the task already changes, and ask "
+        "the user before refactoring code outside them."
     )
 
 

package/hooks/hooks.json

@@ -69,7 +69,7 @@
         "hooks": [
           {
             "type": "command",
-            "command": "python3 ${CLAUDE_PLUGIN_ROOT}/hooks/session/style_reminder_prompt.py --include-skill-reminder",
+            "command": "python3 ${CLAUDE_PLUGIN_ROOT}/hooks/session/skill_loaded_reminder.py",
             "timeout": 10
           }
         ]

package/hooks/hooks_constants/issue_tracker_session_starter_constants.py

@@ -18,6 +18,5 @@ ISSUE_TRACKER_STARTER_TIMEOUT_MILLISECONDS: int = 50
 ISSUE_TRACKER_SESSION_START_DIRECTIVE: str = (
     "SessionStart issue-tracker opt-in is active for this registered repository. "
     "Load the issue-tracker skill when the session needs one GitHub issue action "
-    "(open epic, file sub-issue, update status, refresh checklist, close sub-issue). "
-    "Manual issue-tracker agent spawn remains available unchanged."
+    "(open epic, file sub-issue, update status, refresh checklist, close sub-issue)."
 )

package/hooks/hooks_constants/orchestrator_auto_starter_constants.py

@@ -18,6 +18,5 @@ ORCHESTRATOR_STARTER_TIMEOUT_MILLISECONDS: int = 50
 ORCHESTRATOR_SESSION_START_DIRECTIVE: str = (
     "SessionStart orchestrator opt-in is active. Load the orchestrator skill "
     "and run in executor-advisor mode for this session's multi-step work. "
-    "Do not invent a second orchestrator path; follow the skill. Manual "
-    "/orchestrator remains available unchanged."
+    "Do not invent a second orchestrator path; follow the skill."
 )

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "claude-dev-env",
-    "version": "8.15.0",
+    "version": "8.15.1",
     "description": "Claude Code development standards — rules, hooks, agents, commands, and skills",
     "type": "module",
     "bin": {

package/rules/explore-thoroughly.md

@@ -9,7 +9,6 @@ Note: This deliberately chooses exploration depth over the "commit and execute q
 - Read the relevant files. Understand what exists before proposing what to change.
 - Map the existing patterns: naming conventions, file organization, architectural decisions.
 - Identify constraints that could invalidate an approach before investing effort in it.
-- For unfamiliar codebases or high-stakes changes, invest more time exploring than feels necessary.
 
 ## Exploration scales with risk
 

package/rules/filesystem-search.md

@@ -47,5 +47,5 @@ Issue one shell search at a time when the walk is large. Parallel full-tree sear
 
 ## Search controls
 
-- No hook denies a walk from an unscoped root. `unscoped_search_blocker` went with the blocking policy hooks in `0f21faf8e`, which left the Bash PreToolUse roster empty. The scope invariant above is guidance a reader follows.
+- No hook denies a walk from an unscoped root. The scope invariant above is guidance a reader follows.
 - For Everything searches that use a project name, run `python "${CLAUDE_SKILL_DIR}/scripts/everything_search.py" <project-name> <search arguments>`. An exact project name becomes its path from `~/.claude/project-paths.json`. The command then starts `es.exe` without a shell. `scripts/setup_project_paths.py` writes the registry.

package/rules/git-workflow.md

@@ -15,7 +15,7 @@ User-level rule: applies to **every** git repo that uses GitHub with `gh`. Small
 
 ## Pull request submission rules
 
-**ALWAYS create PRs as DRAFT:** Use `gh pr create --draft` for ALL PRs
+**Create every pull request as a draft.** Use `gh pr create --draft`.
 
 **A release bot's PR body is machine input. Leave it alone.** Release automation reads
 back the body of its own merged pull request to decide it owns that merge. Rewriting the
@@ -47,7 +47,7 @@ untouched.
 
 ## Git golden rules
 
-1. **DRAFT BEFORE PUSH**: When pushing ANYTHING to a PR, it MUST be in draft state first
+1. **Draft before push.** Put a pull request in draft state before you push to it.
    - Before push: `gh pr ready --undo`
    - After review approved: `gh pr ready`
 
@@ -101,7 +101,7 @@ before you ask anyone to read or merge it.
 
 ## Never commit working documents or images
 
-**NEVER commit these files to the repo:**
+**Keep these files out of the repository:**
 
 | Pattern | Reason |
 |---------|--------|
@@ -117,10 +117,10 @@ An image a PR needs as visual evidence is not an exception to that row. Upload i
 **When this applies:** GitHub PR review feedback on a branch you are fixing.
 
 1. Fetch every reviewer comment before making any fix.
-2. Create a TodoWrite checklist with one item per comment.
+2. Create a checklist in the session's task tool with one item per comment.
 3. Fix systematically, marking each todo complete.
 4. Reply to each comment inline.
 
-Repair only reported findings, then re-verify after every repair.
+Repair only the reported findings.
 
 Every `gh` post in this workflow uses `--body-file` per `gh-cli-conventions.md` and keeps volatile scratch paths out per `durable-post-artifacts.md`. Stage session edits per `re-stage-before-commit.md` before each commit.

package/rules/long-horizon-autonomy.md

@@ -35,11 +35,7 @@ Authority the task already granted stays granted. A later preference about tone,
 
 ## Delegate and keep working
 
-Hand independent subtasks to subagents and keep working while they run; let them run in the background rather than block until each one returns. Reuse a long-lived subagent across related subtasks so its context carries forward and saves repeated reads. Step in when a subagent drifts off track or is missing context.
-
-## Verify your work at intervals
-
-On a long build, set a checkpoint cadence and hold to it. At each interval, check the work so far against the task's stated goals with a fresh-context verifier subagent. A separate verifier in a clean context catches what self-review misses.
+Delegate a large, independent track to a subagent, and keep working while it runs in the background. Do work you can finish in a few tool calls yourself: a few reads, a handful of edits, or a simple check. Reuse a long-lived subagent across related subtasks so its context carries forward and saves repeated reads. Step in when a subagent drifts off track or is missing context.
 
 ## Ground every progress claim
 

package/rules/re-stage-before-commit.md

@@ -12,5 +12,3 @@ Staging covers tracked files you edited. Do not commit untracked files unless th
 - **A preceding `git add` or `git stage`.** `git add <paths> && git commit …` stages the files in its own segment before the commit runs.
 
 A `--amend` carries the same risk. An amend records the staged snapshot too, so an unstaged session edit is dropped the same way a plain commit drops it.
-
-`session_file_edit_tracker` (PostToolUse) still records each Write/Edit/MultiEdit path, and `session_edit_tracker_cleanup` (SessionStart, SessionEnd) still clears the session's tracker. No gate reads that record today, so it serves as history rather than a precondition.

package/rules/shell-invocation.md

@@ -18,7 +18,7 @@ When a script file's literal body needs `$(...)`, author it with the Write tool,
 
 ## Enforcement
 
-No PreToolUse hook denies a Bash command. Commit `0f21faf8e` retired the blocking policy hooks and left the Bash PreToolUse roster empty. `shell_substitution_blocker.py` was one of them. The substitution constraint above is guidance a reader follows, and a permission prompt on a wrapped command is the signal that one slipped through.
+No PreToolUse hook denies a Bash command. The substitution constraint above is guidance a reader follows, and a permission prompt on a wrapped command is the signal that one slipped through.
 
 One PreToolUse hook does run on a Bash command, and it only rewrites. `blocking/msys_rev_path_rewriter.py` (PreToolUse on Bash, hosted by `bash_pre_tool_use_dispatcher`) reads a git command before it runs and keeps Git Bash from converting a `<rev>:<path>` argument. That roster holds this one hook, and a test asserts its whole content, so a blocking hook added beside it fails the suite.
 

package/scripts/dev_env_scripts_constants/grok_worker_constants.py

@@ -616,9 +616,15 @@ ADVISOR_CLI_RESUME_FLAG: str = "--resume"
 ADVISOR_BIND_PROMPT_TEMPLATE: str = (
     "You are the unique worker advisor for role {role_name}. "
     "Answer only. Open with exactly one of: ENDORSE | CORRECTION | PLAN | STOP. "
-    "Pre-dispatch: ENDORSE this worker assignment if safe, else CORRECTION."
+    "Pre-dispatch consult: the full worker assignment follows. "
+    "Hunt for missing requirements, wrong assumptions, unsafe actions, and unclear "
+    "completion criteria. ENDORSE when the worker can start as written; "
+    "otherwise open with CORRECTION, PLAN, or STOP and name the fix.\n"
+    "(Advisor: please keep your guidance under 80 words, I need a focused "
+    "starting point.)\n\n"
+    "Worker assignment:\n{assignment_text}"
 )
-"""Prompt body used for the pre-dispatch advisor bind."""
+"""Prompt body for the pre-dispatch advisor consult; carries the assembled worker prompt."""
 
 ADVISOR_VERDICT_PROMPT_TEMPLATE: str = (
     "You are the unique worker advisor for role {role_name}, session {session_id}. "

package/scripts/policy_lint/config/constants.py

@@ -7,6 +7,7 @@ UTF8_ENCODING = "utf-8"
 PATH_SEPARATOR = "/"
 ARCHIVED_SKILLS_DIRECTORY_NAME = "skill-archive"
 ARCHIVED_HOOKS_DIRECTORY_NAME = "hooks-archived"
+ARCHIVED_AUDIT_RUBRICS_DIRECTORY_NAME = "audit-rubrics-archived"
 NUL_BYTE = b"\0"
 ALL_GIT_ROOT_ARGUMENTS = ("rev-parse", "--show-toplevel")
 ALL_GIT_HEAD_ARGUMENTS = ("rev-parse", "--verify", "HEAD")

package/scripts/policy_lint/engine.py

@@ -4,6 +4,7 @@ from collections.abc import Iterable, Sequence
 from pathlib import Path, PurePosixPath
 
 from .config.constants import (
+    ARCHIVED_AUDIT_RUBRICS_DIRECTORY_NAME,
     ARCHIVED_HOOKS_DIRECTORY_NAME,
     ARCHIVED_SKILLS_DIRECTORY_NAME,
 )
@@ -30,7 +31,8 @@ def lint(
     """Run the selected rules on one source selection.
 
     Repository and diff selections check active code. File and editor
-    selections include archived skills and archived hooks.
+    selections include archived skills, archived hooks, and archived
+    audit rubrics.
 
     Args:
         request: Repository root, source, and rule sets.
@@ -61,6 +63,7 @@ def _is_runtime_path(path: PurePosixPath) -> bool:
     return (
         path.parts[0] != ARCHIVED_SKILLS_DIRECTORY_NAME
         and ARCHIVED_HOOKS_DIRECTORY_NAME not in path.parts
+        and ARCHIVED_AUDIT_RUBRICS_DIRECTORY_NAME not in path.parts
     )
 
 
@@ -252,7 +255,7 @@ def _scoped_diagnostics(
     all_diagnostics: Iterable[Diagnostic],
     document: Document,
 ) -> tuple[Diagnostic, ...]:
-    if document.changed_lines is None or not document.changed_lines:
+    if document.changed_lines is None:
         return tuple(all_diagnostics)
     return tuple(
         each_diagnostic

package/scripts/repository_checks/config/constants.py

@@ -118,7 +118,7 @@ ALL_TRACKED_SECRET_EXACT_EXEMPTIONS: frozenset[tuple[str, str, str]] = frozenset
             "f9e409c5c16ecfdd5883da4f3ad045b4af5677740c9e703ad2d47ba231718350",
         ),
         (
-            "packages/claude-dev-env/audit-rubrics/prompts/category-n-test-name-scenario-verifier.md",
+            "packages/claude-dev-env/audit-rubrics-archived/prompts/category-n-test-name-scenario-verifier.md",
             "home-path",
             "6b98ff7f0398f06452f19c4a93f3d69186401dab40aaa7077b394ef59b3e730e",
         ),

package/scripts/spawn_grok_batch.py

@@ -290,6 +290,7 @@ def bind_unique_worker_advisor(
     *,
     advisor_spec: AdvisorSpec,
     role_name: str,
+    assignment_text: str,
     all_used_session_ids: set[str],
 ) -> tuple[str, str]:
     """Bind one unique advisor session for a worker; refuse duplicates and sentinels.
@@ -303,6 +304,7 @@ def bind_unique_worker_advisor(
     Args:
         advisor_spec: Lead-supplied launcher/model/effort.
         role_name: Worker role used in the bind prompt.
+        assignment_text: Assembled worker prompt the advisor reviews before dispatch.
         all_used_session_ids: Sessions already issued in this batch.
 
     Returns:
@@ -317,7 +319,9 @@ def bind_unique_worker_advisor(
         )
     if advisor_spec.launcher == PENDING_BIND_SENTINEL:
         raise ValueError(f"advisor.launcher is {PENDING_BIND_SENTINEL}")
-    prompt_text = ADVISOR_BIND_PROMPT_TEMPLATE.format(role_name=role_name)
+    prompt_text = ADVISOR_BIND_PROMPT_TEMPLATE.format(
+        role_name=role_name, assignment_text=assignment_text
+    )
     session_id, advisor_body_text, returncode = batch_invoke_advisor(
         launcher=advisor_spec.launcher,
         model=advisor_spec.model,
@@ -898,6 +902,7 @@ def _bind_worker_advisor_session(
     *,
     advisor_spec: AdvisorSpec,
     role_name: str,
+    assignment_text: str,
     all_used_session_ids: set[str],
     session_id_lock: object | None,
 ) -> tuple[str, str]:
@@ -906,11 +911,13 @@ def _bind_worker_advisor_session(
             return bind_unique_worker_advisor(
                 advisor_spec=advisor_spec,
                 role_name=role_name,
+                assignment_text=assignment_text,
                 all_used_session_ids=all_used_session_ids,
             )
     return bind_unique_worker_advisor(
         advisor_spec=advisor_spec,
         role_name=role_name,
+        assignment_text=assignment_text,
         all_used_session_ids=all_used_session_ids,
     )
 
@@ -990,6 +997,10 @@ def _maybe_bind_advisor_for_worker(
     advisor_session_id, pre_signal = _bind_worker_advisor_session(
         advisor_spec=advisor_spec,
         role_name=worker_spec.role_name,
+        assignment_text=assemble_worker_prompt(
+            all_prompt_part_paths=worker_spec.all_prompt_part_paths,
+            tool_profile=worker_spec.tool_profile,
+        ),
         all_used_session_ids=used_ids,
         session_id_lock=session_id_lock,
     )

package/scripts/test_spawn_grok_batch.py

@@ -1514,6 +1514,7 @@ def test_bind_unique_worker_advisor_rejects_placeholder(
         batch.bind_unique_worker_advisor(
             advisor_spec=batch.AdvisorSpec(launcher=batch.DEFAULT_ADVISOR_LAUNCHER_PLACEHOLDER),
             role_name="lens",
+            assignment_text="assignment",
             all_used_session_ids=set(),
         )
 
@@ -1528,12 +1529,33 @@ def test_bind_unique_worker_advisor_returns_session(
     session_id, signal = batch.bind_unique_worker_advisor(
         advisor_spec=batch.AdvisorSpec(launcher="fixture-advisor-launcher"),
         role_name="lens",
+        assignment_text="assignment",
         all_used_session_ids=set(),
     )
     assert session_id == "sess-unique-1"
     assert signal == "ENDORSE"
 
 
+def test_bind_unique_worker_advisor_prompt_carries_assignment_text(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    all_captured_prompts: list[object] = []
+
+    def fake(**kwargs: object) -> tuple[str | None, str, int]:
+        all_captured_prompts.append(kwargs["prompt_text"])
+        return "sess-assignment-1", "ENDORSE\nok", 0
+
+    monkeypatch.setattr(batch, "batch_invoke_advisor", fake)
+    batch.bind_unique_worker_advisor(
+        advisor_spec=batch.AdvisorSpec(launcher="fixture-advisor-launcher"),
+        role_name="lens",
+        assignment_text="Audit module alpha for unused imports.",
+        all_used_session_ids=set(),
+    )
+    assert len(all_captured_prompts) == 1
+    assert "Audit module alpha for unused imports." in str(all_captured_prompts[0])
+
+
 def test_obtain_advisor_completion_verdict_endorses(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:

package/scripts/tests/test_banned_prose_words.py

@@ -86,7 +86,7 @@ def test_governed_paths_lists_the_shipped_surfaces(tmp_path: Path) -> None:
 def test_governed_paths_cover_instruction_surfaces_and_skip_archives() -> None:
     assert governs_path("rules/git-workflow.md")
     assert governs_path(".agents/skills/eli5/SKILL.md")
-    assert governs_path("audit-rubrics/prompts/category-e-dead-code.md")
+    assert governs_path("audit-rubrics/prompts/category-o-docstring-vs-impl-drift.md")
     assert governs_path("system-prompts/software-engineer.xml")
     assert not governs_path("rules-archived/manifest.md")
     assert not governs_path("scripts/banned_prose_words.py")

package/scripts/tests/test_policy_lint_rules_archive.py

@@ -134,3 +134,19 @@ def test_archived_hook_moves_leave_the_automatic_checks(tmp_path: Path) -> None:
     assert runtime_set.documents == ()
     assert runtime_set.deleted_paths == (live_hook_path,)
     assert runtime_set.renamed_paths == ()
+
+
+def test_archived_audit_rubric_moves_leave_the_automatic_checks(tmp_path: Path) -> None:
+    live_rubric_path = PurePosixPath("packages/tool/audit-rubrics/prompts/category-k.md")
+    archived_rubric_path = PurePosixPath("packages/tool/audit-rubrics-archived/prompts/category-k.md")
+    document_set = DocumentSet(
+        (Document.from_text(archived_rubric_path, "archived"),),
+        SelectionKind.BASE,
+        tmp_path,
+        (),
+        ((live_rubric_path, archived_rubric_path),),
+    )
+    runtime_set = _runtime_document_set(document_set)
+    assert runtime_set.documents == ()
+    assert runtime_set.deleted_paths == (live_rubric_path,)
+    assert runtime_set.renamed_paths == ()

package/scripts/tests/test_policy_lint_rules_engine.py

@@ -209,7 +209,7 @@ def test_changed_line_filter_drops_other_located_line(tmp_path: Path) -> None:
     assert lint_report.diagnostics == ()
 
 
-def test_lint_keeps_located_findings_when_changed_lines_are_empty(
+def test_lint_drops_located_findings_when_a_change_only_deletes_lines(
     tmp_path: Path,
 ) -> None:
     repository_root = tmp_path.resolve()
@@ -233,7 +233,7 @@ def test_lint_keeps_located_findings_when_changed_lines_are_empty(
         (document,),
         (_changed_document_rule("probe", collect_diagnostics),),
     )
-    assert len(lint_report.diagnostics) == 1
+    assert lint_report.diagnostics == ()
 
 
 def test_code_rule_diagnostics_includes_comment_policy_without_prior_text(

package/scripts/tests/test_rules.py

@@ -25,10 +25,10 @@ def _comment_policy_surfaces(package_root: Path) -> tuple[Path, ...]:
         package_root / "system-prompts" / "software-engineer.xml",
         package_root / ".agents" / "agents-archived" / "clean-coder.md",
         package_root / ".agents" / "agents-archived" / "code-quality-agent.md",
-        package_root / "audit-rubrics" / "category_rubrics" / "category-j-code-rules-compliance.md",
-        package_root / "audit-rubrics" / "prompts" / "category-j-code-rules-compliance.md",
-        package_root / "audit-rubrics" / "category_rubrics" / "category-l-behavior-equivalence.md",
-        package_root / "audit-rubrics" / "prompts" / "category-l-behavior-equivalence.md",
+        package_root / "audit-rubrics-archived" / "category_rubrics" / "category-j-code-rules-compliance.md",
+        package_root / "audit-rubrics-archived" / "prompts" / "category-j-code-rules-compliance.md",
+        package_root / "audit-rubrics-archived" / "category_rubrics" / "category-l-behavior-equivalence.md",
+        package_root / "audit-rubrics-archived" / "prompts" / "category-l-behavior-equivalence.md",
         package_root / ".agents" / "skills" / "grok-spawn" / "reference" / "worker-briefs.md",
         package_root.parent.parent / ".github" / "copilot-instructions.md",
     )
@@ -96,17 +96,17 @@ def test_comment_guidance_reaches_installed_instruction_surfaces() -> None:
 def _category_l_policy_texts() -> tuple[str, str, str]:
     category_l_rubric = (
         _PACKAGE_ROOT
-        / "audit-rubrics"
+        / "audit-rubrics-archived"
         / "category_rubrics"
         / "category-l-behavior-equivalence.md"
     )
     category_l_prompt = (
         _PACKAGE_ROOT
-        / "audit-rubrics"
+        / "audit-rubrics-archived"
         / "prompts"
         / "category-l-behavior-equivalence.md"
     )
-    audit_categories = _PACKAGE_ROOT / "audit-rubrics" / "audit-categories.json"
+    audit_categories = _PACKAGE_ROOT / "audit-rubrics-archived" / "audit-categories.json"
     return tuple(
         each_path.read_text(encoding="utf-8").lower()
         for each_path in (category_l_rubric, category_l_prompt, audit_categories)
@@ -117,7 +117,7 @@ def _comment_policy_summary_texts() -> tuple[str, str, str]:
     clean_coder = _PACKAGE_ROOT / ".agents" / "agents-archived" / "clean-coder.md"
     category_j = (
         _PACKAGE_ROOT
-        / "audit-rubrics"
+        / "audit-rubrics-archived"
         / "category_rubrics"
         / "category-j-code-rules-compliance.md"
     )
@@ -145,7 +145,7 @@ def test_comment_policy_removes_directive_justification_guidance() -> None:
     ).read_text(encoding="utf-8").lower()
     category_e_prompt = (
         _PACKAGE_ROOT
-        / "audit-rubrics"
+        / "audit-rubrics-archived"
         / "prompts"
         / "category-e-dead-code.md"
     ).read_text(encoding="utf-8").lower()

package/system-prompts/software-engineer.xml

@@ -67,17 +67,7 @@
   Reserve user questions for preferences, missing context the user holds, judgment calls, and
   scope decisions.
 
-  Three anti-hallucination constraints are always active. First, acknowledge the gap directly
-  — "I lack data on this" — when a credible source for a claim is absent; guessing and
-  inferring are replaced by honest acknowledgment. Second, every recommendation, claim, or piece of advice cites a
-  specific source: a file in the current project, an external URL found via web search, a
-  named expert or paper, or official documentation. Third, when working from documents, extract
-  the text first before analyzing — ground responses in direct quotes, and reference
-  the quote when making a point.
-
-  Creative thinking, brainstorming, and novel ideas are exempt from citation requirements;
-  synthesizing across sources to reach new conclusions is encouraged provided the inputs are
-  grounded.
+  Source every claim as rules/research-mode.md describes.
 </investigation>
 
 <problem_solving>
@@ -145,8 +135,8 @@
 </code_quality>
 
 <behavior_protocol>
-  Apply an outside-in BDD reasoning protocol before writing any code. Outside-in BDD is the
-  required development protocol; apply it in place of any Red-Green-Refactor heuristic. The
+  Apply an outside-in BDD reasoning protocol before writing any code. BDD is the outer
+  process, and red-green-refactor is the inner loop for each formulated behavior. The
   full anti-pattern catalog, Example Mapping algorithm, and solo-developer guidance live in
   the long-form BDD references (`~/.claude/docs/BDD_DISCOVERY_PROTOCOL.md`,
   `~/.claude/docs/BDD_SCENARIO_QUALITY.md`); this section carries the core four-phase sequence.
@@ -178,9 +168,9 @@
   where untrusted data enters the system. Trust internal code and framework guarantees within
   those boundaries; internal call sites operate on data already validated at the entry point.
 
-  Create abstractions and shared helpers when a pattern recurs three or more times. A single
-  occurrence warrants inline code; two occurrences are worth watching; three occurrences
-  justify extraction. Implement precisely what the task requires at its current scale and
+  Add an abstraction at the commit that introduces its second concrete implementation, the
+  house setting in rules/code-standards.md. Implement precisely what the task requires at its
+  current scale and
   complete it fully.
 
   Apply backwards compatibility conventions — underscore-prefixed variable renaming, type
@@ -216,11 +206,6 @@
   success looks like; and that the task is unambiguous enough to delegate. When any of those
   is unknown, investigate first using Read, Glob, or Grep for code search — or ask the user.
 
-  After confirming context, use the /prompt-generator skill to produce a structured prompt.
-  Feed it the task description and goal, target files and directories, constraints and
-  boundaries, expected output format, and acceptance criteria. Pass the skill's output as the
-  agent's prompt parameter.
-
   When multiple tool calls are mutually independent, make all calls in a single response.
   Sequence calls only when a later call requires an earlier call's result. Reading three files:
   call all three Read operations at once. Running independent searches: launch all Grep
@@ -234,7 +219,7 @@
   Fetch all reviewer comments before making any fixes. Obtain the full comment set from the
   PR using `gh pr view --comments`.
 
-  Create a TodoWrite checklist with one item per comment.
+  Create a checklist in the session's task tool with one item per comment.
 
   Fix comments systematically, marking each todo complete as it is resolved.
 
@@ -262,8 +247,9 @@
 </documentation>
 
 <context7>
-  For all work involving libraries, frameworks, SDKs, or APIs, fetch current documentation
-  via Context7 MCP before writing code. Training data may be stale; fetched docs are
+  When the Context7 MCP server is connected, fetch current documentation before you write code
+  against a library, framework, SDK, or API whose interface you have not checked this session.
+  Training data may be stale; fetched docs are
   authoritative. Apply this to setup questions, API references, version migration,
   library-specific debugging, and CLI tool usage — including well-known libraries such as
   React, Next.js, Prisma, Express, Django, and Spring Boot.
@@ -277,8 +263,9 @@
 </context7>
 
 <obsidian_vault>
-  Use the Obsidian vault path configured for your workspace (session reports, decisions, and
-  research documents typically live there across projects). Search it before starting substantive project work.
+  When an Obsidian MCP server is connected, use the vault path configured for your workspace
+  (session reports, decisions, and research documents typically live there across projects).
+  Search it before starting substantive project work.
 
   Available MCP tools:
   - mcp__obsidian__search_notes — search by content or frontmatter (searchFrontmatter: true)
@@ -356,7 +343,7 @@
 
   Reserve Bash for system commands and terminal operations that require shell execution.
 
-  Track work with TodoWrite and mark each task complete as soon as it is finished.
+  Track work with the session's task tool and mark each task complete as soon as it is finished.
 </tool_usage>
 
 <cleanup>
@@ -375,9 +362,7 @@
 <communication_contract>
   Visible output is concise by default. The first progress update is one sentence.
   Later updates mark only important discoveries or direction changes. The final
-  response begins with the outcome. When thinking is disabled, speak one brief
-  natural-language sentence before a tool call that needs context, state when no
-  fitting tool exists, and never emit internal-system XML in visible output.
+  response begins with the outcome.
   Language policy: rules/asd-ste100-language.md. This section owns visible progress,
   tool narration, and outcome-first final structure.
 </communication_contract>