chore(hooks): remove automatic test runs

config/personas/quality/references/checks.conf

@@ -8,10 +8,8 @@ IMMEDIATE="uv run python scripts/validate_schemas.py"
 # Before each commit: affected integration tests and a small eval smoke set.
 CHECKPOINT="uv run pytest -m integration -q && uv run python -m evals.smoke"
 
-# Tests the checkpoint runs: related (default) runs only the tests the commit's
-# changes reach, leaving the full suite to CI; full runs each affected
-# package's whole suite. Use full only when the suite takes seconds.
-# CHECKPOINT_TESTS=full
+# Hooks do not automatically run tests. Set CHECKPOINT explicitly for any
+# tests this repo needs before an agent commit, or use your own pre-commit hook.
 
 # Nightly CI (quality.yml schedule): E2E, broad model matrix, load tests, the
 # full eval suite.

docs/hooks.md

@@ -24,19 +24,20 @@ Each check runs at the cheapest loop that still catches the problem:
 
 | Loop | What runs | When |
 |---|---|---|
-| Immediate | Per edit: format, lint and type check on the file. Per turn: related unit tests, types, layer rules, duplication, and the project's `IMMEDIATE` command | Every agent iteration |
-| Local checkpoint | The tests the commit's changes reach (each affected package's whole suite with `CHECKPOINT_TESTS=full`), the same lint/type/layer checks, and the project's `CHECKPOINT` command (integration tests, an eval smoke set) | When the agent runs `git commit`; a failure blocks the commit |
+| Immediate | Per edit: format, lint and type check on the file. Per turn: types, layer rules, duplication, and the project's `IMMEDIATE` command | Every agent iteration |
+| Local checkpoint | Lint/type/layer checks, synthesis, review, and the project's explicit `CHECKPOINT` command | When the agent runs `git commit`; a failure blocks the commit |
 | PR | Full tests, semgrep, dependency audits, jscpd threshold (the `quality` skill's CI workflow) | Before human review |
 | Merge/nightly | The project's `NIGHTLY` command: E2E, broad model matrix, load tests, full eval suite | Scheduled CI job |
 | Production | Traces, user failures, latency and cost, sampled quality evals | Continuously; see the `sre` skill |
 
 Projects declare their own commands in `.macols/checks.conf` (template in
 `config/personas/quality/references/checks.conf`). The hooks read it as data and
 run the commands from the repo root in trusted projects only. The
-checkpoint uses the turn-end test selection over the whole uncommitted change
-and leaves the full suite to CI; set `CHECKPOINT_TESTS=full` there to run each
-affected package's whole suite instead. The checkpoint
-skips when the tree matches the last passing checkpoint, when the command
+hooks never automatically run tests. Repos own test execution through their
+own pre-commit/CI workflow or an explicit `CHECKPOINT` command. Legacy
+`CHECKPOINT_TESTS` and test-scope environment settings do not enable automatic
+tests. Existing `IMMEDIATE` or `CHECKPOINT` commands still run as configured.
+The checkpoint skips when the tree matches the last passing checkpoint, when the command
 passes `--no-verify`, or with `MACOLS_CHECKPOINT=off`.
 
 ## What Runs
@@ -50,13 +51,10 @@ Every tool gets the findings in its own model-visible format
   For Go that means a compile or type error in the edited package shows up
   with the edit, not at turn end; go vet builds the package, so it runs in
   trusted projects only.
-- post-task: at the end of a turn, tests related to the changed files, lint
-  and strict types, golangci-lint, jscpd duplication, and layer rules
-  (import-linter, dependency-cruiser) when the project configures them. Test
-  selection: pytest-testmon when the project's venv has it (coverage-based),
-  otherwise `test_<module>.py` by name; `vitest related` / `jest
-  --findRelatedTests` by import graph (any other `npm test` waits for the
-  checkpoint, since it may be E2E); `go test` on the changed packages. When
+- post-task: at the end of a turn, lint and strict types, golangci-lint,
+  jscpd duplication, CDK synthesis, and layer rules (import-linter,
+  dependency-cruiser) when the project configures them. Built-in Python,
+  Node, Flutter and Go tests do not run. When
   something fails, the agent gets one more step with the findings and a fix
   instruction. It only re-runs when the tree changed since its last run, so it
   can't loop and Q&A turns stay fast (under a second on small projects).
@@ -142,20 +140,16 @@ first run). Pass `--no-cli` to skip the whole step.
 
 Every turn-end and checkpoint run appends one JSON line per check to
 `.git/macols-checks.jsonl` (shared by worktrees, capped at 10k lines):
-duration, pass or fail, and which loop ran it. At the checkpoint, failing
-Python tests that the name-based turn-end selection wouldn't have run are
-recorded as escapes.
+duration, pass or fail, and which loop ran it. Historical test escapes remain readable in older logs.
 
 ```bash
 macols-check-stats             # per loop and check: runs, total/p95 time, fail %, fail streak, escapes
 macols-check-stats --days 7
 macols-check-stats --escapes   # which tests escaped the fast path
 ```
 
-Read it as: a slow check that never fails belongs in CI, a long fail streak
-means the agent is going round in circles, and escapes mean the fast path
-needs pytest-testmon or better-named tests. `MACOLS_CHECK_LOG=off` stops the
-logging.
+A slow check that never fails is a candidate for CI. A long fail streak
+means the agent is going round in circles. `MACOLS_CHECK_LOG=off` stops logging.
 
 ## Trusted Projects
 
@@ -190,16 +184,12 @@ workflow. Existing layer breaks are approved explicitly: a
 
 | Variable | Effect |
 |---|---|
-| `MACOLS_PYTEST_SCOPE=changed\|full\|off` | Python test scope at turn end (default `changed`) |
-| `MACOLS_TESTMON=off` | Ignore pytest-testmon and select Python tests by name |
-| `MACOLS_GO_TEST_SCOPE=changed\|module` | Go tests at turn end: only changed packages (default) or the whole changed module via the test cache |
 | `MACOLS_DEPLOY_DIFF=off` | Don't run `cdk diff` before a deploy prompt |
 | `MACOLS_DEPLOY_DIFF_TIMEOUT` | Seconds the pre-deploy `cdk diff` may take (default 240) |
 | `MACOLS_CHECK_LOG=off` | Don't log check timings |
 | `MACOLS_DUPLICATION=off` | Skip the jscpd duplication check |
 | `MACOLS_DUP_MIN_LINES` | Minimum clone size jscpd reports (default 10) |
 | `MACOLS_MAX_FILE_LINES` | File-length limit (default 500, `0` to turn off) |
-| `MACOLS_GO_RACE=1` | Run Go tests with the race detector |
 | `MACOLS_SEMGREP=1` | Run semgrep locally |
 | `MACOLS_CHECKS_VERBOSE=1` | Also print notes such as "pytest not installed" |
 | `MACOLS_HOOKS_DIR` | Where the hook scripts live |

hooks/checks/lang_checks.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 #
 # lang_checks.sh — the remaining per-language batteries: cdk synth, Flutter
-# (scoped flutter test + dart analyze), Go (golangci-lint/go vet + go test),
+# (dart analyze), Go (golangci-lint/go vet),
 # and shellcheck for scripts.
 #
 # Sourced by post_task.sh AFTER common.sh and its helpers (add_critical_issue,
@@ -100,10 +100,7 @@ _changed_go_packages() {
 }
 
 # golangci-lint (the project's .golangci.yml: dupl, gocyclo, funlen, depguard…)
-# or go vet when it isn't installed, on the changed packages; then go test on
-# the changed packages (./... per module at the checkpoint).
-# MACOLS_GO_TEST_SCOPE=module widens turn-end tests to the whole module, which
-# also catches dependents (cheap when the test cache can replay them).
+# or go vet when it isn't installed, on the changed packages.
 run_go_checks() {
     command -v go &> /dev/null || { add_warning "go not installed - skipping Go checks"; return 0; }
     local pkgs
@@ -122,13 +119,6 @@ run_go_checks() {
             out=$(cd "$mod" && go vet "${dirs[@]}" 2>&1) || ec=$?
             report_check_result "go vet ($label)" "$ec" "$out" "^.+\.go:[0-9]+(:[0-9]+)?: |^vet: " "issues"
         fi
-        ec=0
-        local -a race=()
-        [ "${MACOLS_GO_RACE:-0}" = "1" ] && race=(-race)
-        local -a test_targets=(./...)
-        [ "${MACOLS_GO_TEST_SCOPE:-changed}" = "changed" ] && ! full_suite && test_targets=("${dirs[@]}")
-        out=$(cd "$mod" && ${TIMEOUT_CMD:+$TIMEOUT_CMD $MAX_TEST_TIME} go test ${race[@]+"${race[@]}"} "${test_targets[@]}" 2>&1) || ec=$?
-        [ "$ec" -ne 0 ] && report_test_failure "go test ($label)" "$ec" "$out"
     done
     return 0
 }

hooks/checks/post_task.sh

@@ -25,10 +25,7 @@
 # per-turn loop (semgrep, pip-audit/npm audit/govulncheck — see the `quality`
 # skill's CI template).
 #
-#   Tests         pytest (scoped, or pytest-testmon when installed),
-#                 jest/vitest related tests, flutter test (scoped), go test
-#                 over changed modules (the test cache skips the rest),
-#                 cdk synth
+#   Synthesis     cdk synth
 #   Lint          ruff, eslint, dart analyze, golangci-lint (or go vet),
 #                 and shellcheck, each with the project's own limits
 #   Types         pyright or mypy (whichever the project configures),
@@ -42,9 +39,8 @@
 # Only findings are recorded as CRITICAL_ISSUES, each with a fix instruction.
 # WARNINGS hold notes (tool missing, nothing to test) for verbose runs.
 #
-# Switches: MACOLS_PYTEST_SCOPE=changed|full|off, MACOLS_TESTMON=off,
-# MACOLS_DUPLICATION=off, MACOLS_SEMGREP=1 (opt back in to a local semgrep
-# scan), MACOLS_GO_RACE=1, MACOLS_GO_TEST_SCOPE=changed|module,
+# Tests run only through explicit project commands, never automatically.
+# Switches: MACOLS_DUPLICATION=off, MACOLS_SEMGREP=1,
 # MACOLS_CHECK_LOG=off (see log_check_run in common.sh),
 # MACOLS_LGREVIEW=off (and its BLOCK/TIMEOUT knobs — lgtmaybe_review.sh).
 #
@@ -208,14 +204,14 @@ run_post_task_checks() {
 
     local -a checks=()
     if [ "$has_python" = "true" ]; then
-        checks+=(run_python_tests run_ruff_check run_python_typecheck run_import_linter)
+        checks+=(run_ruff_check run_python_typecheck run_import_linter)
     fi
     if [ "$has_node" = "true" ]; then
-        checks+=(run_node_tests run_eslint_check run_tsc_check run_dependency_cruiser)
+        checks+=(run_eslint_check run_tsc_check run_dependency_cruiser)
     fi
     [ "$has_cdk" = "true" ] && checks+=(run_cdk_tests)
     if [ "$has_flutter" = "true" ]; then
-        checks+=(run_flutter_tests run_dart_analyze)
+        checks+=(run_dart_analyze)
     fi
     [ "$has_go" = "true" ] && checks+=(run_go_checks)
     checks+=(run_shellcheck run_duplication_check run_file_length_check run_semgrep_scan)

hooks/post_task_hook.sh

@@ -17,14 +17,12 @@
 #   changed nothing is allowed to stop, so the hook cannot loop.
 # - Never when the tool says this stop is already a hook-driven continuation
 #   (stop_hook_active) — one nudge per turn, then the user decides.
-# - pytest/jest/vitest/go test are scoped to what the changed files reach
-#   (pytest-testmon or file names for Python, the test cache for Go). The full
-#   suite is the commit checkpoint's and CI's job.
-#   MACOLS_PYTEST_SCOPE=full|changed|off overrides the Python scope.
+# - Built-in tests never run. Repos own test execution through explicit
+#   project commands or their own pre-commit/CI workflow.
 # - Every run appends per-check timings to .git/macols-checks.jsonl
 #   (bin/macols-check-stats summarises them).
 #
-# Only critical findings (failing tests, lint/type/gate errors) are sent to the
+# Only critical findings (failing project commands, lint/type/gate errors) are sent to the
 # model; notes such as "pytest not installed" print only with
 # MACOLS_CHECKS_VERBOSE=1.
 #
@@ -78,7 +76,7 @@ fi
 
 TRUST_NOTE=""
 if [ "${UNTRUSTED_SKIPPED:-0}" = "1" ]; then
-    TRUST_NOTE="Tests, eslint, tsc, mypy, cdk synth, go and layer checks were skipped: $(project_root) is not a trusted project, and those run code the repo controls. To enable them, run $(dirname "$SHARED_HOOKS_ROOT")/bin/macols-trust in the repo, or add its path to $MACOLS_TRUST_FILE."
+    TRUST_NOTE="Project commands, eslint, tsc, mypy, cdk synth, go and layer checks were skipped: $(project_root) is not a trusted project, and those run code the repo controls. To enable them, run $(dirname "$SHARED_HOOKS_ROOT")/bin/macols-trust in the repo, or add its path to $MACOLS_TRUST_FILE."
 fi
 
 if [ ${#CRITICAL_ISSUES[@]} -eq 0 ]; then

hooks/pre_commit_check.sh

@@ -2,8 +2,7 @@
 #
 # pre_commit_check.sh <command-string> — protocol-neutral core of the local
 # checkpoint loop: when the agent runs `git commit`, run the checkpoint
-# battery (the tests the change reaches, or whole suites with
-# CHECKPOINT_TESTS=full; lint, types, layer rules, and the project's
+# battery (lint, types, layer rules, and the project's explicit
 # CHECKPOINT command) before the commit lands.
 #
 # Prints the findings when the checkpoint fails and nothing otherwise, and
@@ -49,9 +48,7 @@ fp=""
 review_ready || fp=$(_check_fingerprint || true)
 [ -n "$fp" ] && [ -f "$fp_file" ] && [ "$(cat "$fp_file")" = "$fp" ] && exit 0
 
-CHECKPOINT_MODE=1 MACOLS_PYTEST_SCOPE=full
-full_suite || MACOLS_PYTEST_SCOPE=changed
-export CHECKPOINT_MODE MACOLS_PYTEST_SCOPE
+export CHECKPOINT_MODE=1
 # shellcheck disable=SC2034  # read by the sourced post_task.sh checks
 MAX_TEST_TIME="${MACOLS_CHECKPOINT_TIMEOUT:-900}"
 run_post_task_checks || exit 0

openspec/changes/remove-hook-test-runs/.openspec.yaml

@@ -0,0 +1,2 @@
+schema: macols
+created: 2026-10-20

openspec/changes/remove-hook-test-runs/design.md

@@ -0,0 +1,9 @@
+# Design
+## Context
+The shared orchestrator dispatches Python, Node and Flutter test checks; Go combines lint and tests in one check. Both turn-end and pre-commit use this battery.
+## Goals / Non-Goals
+Remove automatic test execution. Preserve static checks, synthesis, review, trust boundaries and explicit project commands. Do not provision repo-specific test hooks.
+## Decisions
+Remove test dispatch from the shared orchestrator and the test portion of Go checks. Keep existing standalone Python/Node/Flutter selection helpers outside the automatic battery to avoid unrelated helper and fixture churn. Remove checkpoint scope overrides because the battery no longer selects tests.
+## Risks / Trade-offs
+Tests no longer block completion or commits unless a repo explicitly configures them. Existing IMMEDIATE or CHECKPOINT commands can still run tests by design.

openspec/changes/remove-hook-test-runs/evidence.md

@@ -0,0 +1,20 @@
+# Evidence
+
+## E1. No automatic tests; static checks and configured commands survive
+- **Claim**: Both batteries skip built-in tests, including legacy full-scope settings; Go lint and explicit checkpoint failures remain effective. Trust gates and model-visible findings survive.
+- **Command**: `bash tests/test_hook_loops.sh`
+- **Expected**: All assertions pass using scratch repos and boundary stubs.
+- **Fails before**: The new battery regression detects automatic test dispatch; Go assertions detect go test.
+- **Recorded**: pass, 41/41 assertions; base HEAD 95a50bdf609186b7b0014593bdff3c24bfcdeca3, working-tree changes tested (2026-10-20).
+
+## E2. Shell and anchor constraints
+- **Claim**: C-SHELLCHECK-CLEAN, C-SPEC-ANCHORS-HEALTHY and source-only changes hold.
+- **Command**: `bash -n hooks/checks/post_task.sh hooks/checks/lang_checks.sh hooks/post_task_hook.sh hooks/pre_commit_check.sh tests/test_hook_loops.sh && shellcheck -S warning hooks/checks/post_task.sh hooks/checks/lang_checks.sh hooks/post_task_hook.sh hooks/pre_commit_check.sh tests/test_hook_loops.sh && ./scripts/spec_drift_gate.sh --check`
+- **Expected**: All commands pass; installer and per-edit code unchanged.
+- **Recorded**: pass; base HEAD 95a50bdf609186b7b0014593bdff3c24bfcdeca3, working-tree changes tested (2026-10-20).
+
+## E3. Installer, bundle and change validation
+- **Claim**: Codex wiring remains valid, the quality template is bundled, and spec deltas validate.
+- **Command**: `./tests/verify_install.sh codex && ./scripts/package_claude_desktop_personas.sh --check && openspec validate remove-hook-test-runs --strict && git diff --check`
+- **Expected**: All hard checks pass.
+- **Recorded**: pass; base HEAD 95a50bdf609186b7b0014593bdff3c24bfcdeca3, working-tree changes tested (2026-10-20). Other agents' installer verifiers were not run; shared hook wiring is unchanged.

openspec/changes/remove-hook-test-runs/proposal.md

@@ -0,0 +1,32 @@
+## Why
+Automatic tests can take minutes in large repos and delay every agent reply. Test execution belongs to each repo's chosen pre-commit workflow.
+
+## What Changes
+- Remove built-in Python, Node, Flutter and Go test execution from turn-end and commit batteries.
+- Keep lint, types, structure checks, CDK synthesis, AI review and explicitly configured project commands.
+
+## Capabilities
+### New Capabilities
+None.
+### Modified Capabilities
+- `hook-wiring`: batteries validate code without automatically running tests.
+
+## Impact
+Shared batteries affect every installed agent in place. Repos can explicitly configure CHECKPOINT commands for tests.
+
+## Scope
+### May change
+- `hooks/checks/post_task.sh`, `hooks/checks/lang_checks.sh`, `hooks/post_task_hook.sh`, `hooks/pre_commit_check.sh`
+- `tests/test_hook_loops.sh`, `docs/hooks.md`, `config/personas/quality/references/checks.conf`
+- `openspec/changes/remove-hook-test-runs/**`
+- `dist/macols-personas-claude-plugin.zip`: regenerate the bundled quality template.
+### Must not change
+- Installer wiring, trust gates, per-edit checks, user-owned configuration and unrelated plugins.
+
+## Constraints
+- C-EDIT-SOURCE-NOT-OUTPUT: hooks run from this source checkout.
+- C-TRUSTED-REPO-CODE: explicit project commands retain trust checks.
+- C-HOOKS-MODEL-VISIBLE: preserve finding delivery.
+- C-SHELLCHECK-CLEAN: verify touched shell scripts.
+- C-SPEC-ANCHORS-HEALTHY: keep anchors resolving and propose specification deltas.
+- No new opt-in setting: automatic tests are removed for all repos.

openspec/changes/remove-hook-test-runs/specs/hook-wiring/spec.md

@@ -0,0 +1,42 @@
+## ADDED Requirements
+
+### Requirement: Hook batteries do not automatically execute tests
+Turn-end and commit batteries SHALL retain lint, type, layer, duplication, file-length and synthesis checks without automatically running Python, Node, Flutter or Go tests. Tests SHALL run only when explicitly requested by a repository command or another repository-owned workflow. Legacy test-scope settings SHALL NOT enable automatic test execution.
+
+#### Scenario: Code changes without configured test commands
+- **WHEN** either battery runs in a trusted project with changed Python, Node, Flutter or Go code
+- **THEN** no built-in test runner executes and applicable static checks still run
+
+## MODIFIED Requirements
+
+### Requirement: A local checkpoint runs before commits
+Every tool's PreToolUse(Bash) wiring SHALL pass the command to
+`hooks/pre_commit_check.sh`, which on `git commit` (not `--no-verify`, not
+`MACOLS_CHECKPOINT=off`, trusted projects only) runs lint/type/layer checks,
+synthesis, review and the project's explicit `CHECKPOINT` command from
+`.macols/checks.conf`. A failure SHALL deny the commit with the findings as
+the reason; a pass is remembered by tree fingerprint so an unchanged tree
+is not re-checked. The turn-end battery SHALL run the project's `IMMEDIATE`
+command. No tests SHALL run automatically in either battery.
+<!-- anchor: hook-wiring.checkpoint -->
+
+#### Scenario: The agent commits a failing change
+- **WHEN** the agent runs `git commit` in a trusted repo whose CHECKPOINT command fails
+- **THEN** the commit is denied and the command's findings reach the model
+
+#### Scenario: A commit runs only the tests it reaches
+- **WHEN** the agent runs `git commit` without an explicit test command in CHECKPOINT
+- **THEN** the checkpoint runs its applicable validation checks without invoking tests, including when CHECKPOINT_TESTS=full was previously configured
+
+### Requirement: Every battery run is logged per check
+`run_post_task_checks` SHALL append one JSON line per check, plus a `total`
+line, to `macols-checks.jsonl` in the repo's common git dir (shared by
+worktrees): time, repo, loop (`turn` or `checkpoint`), check, duration in ms
+and `pass`/`fail`. The log SHALL stay bounded (newest 5000 lines once it passes
+10000), `MACOLS_CHECK_LOG=off` SHALL turn it off, and `bin/macols-check-stats`
+SHALL summarise time, failure rate, repeated failures and historical escapes per check.
+<!-- anchor: hook-wiring.check-log -->
+
+#### Scenario: The user asks whether the hooks are worth their time
+- **WHEN** the user runs `macols-check-stats` in a repo after some agent turns
+- **THEN** it prints, per loop and check, runs, total and p95 time, failure rate and historical escaped tests