Enable the content-policy dimension by default

ARCHITECTURE.md

@@ -50,9 +50,8 @@ second model call: guardrails are repo-specific merge rules from `.postil/guardr
 (violations are `kind: guardrail`); content policy reviews human-readable prose only
 (Markdown, comments, docstrings, user-facing strings, PR title/body — never code logic
 or identifiers) against a built-in baseline plus optional `.postil/content-policy.md`
-additions (violations are `kind: contentPolicy`). Content policy is off by default;
-either an explicit `contentPolicy.enabled: true` or the mere presence of
-`.postil/content-policy.md` turns it on, mirroring how guardrails activates.
+additions (violations are `kind: contentPolicy`). Content policy is on by default;
+`contentPolicy.enabled: false` fully disables the baseline and repo additions.
 
 ## Invariants
 

README.md

@@ -99,17 +99,20 @@ the rule it breaks.
 
 ## Content policy
 
-Off by default. Reviews human-readable prose in the diff — Markdown, code comments,
-docstrings, user-facing/log strings, and the PR title/description, never code logic or
-identifiers — for fabricated or contradicted documentation claims, self-contradictions
-the same PR creates, authoring-process narration and AI-authorship residue, leaked
-conversation/transcript text, and (lower severity) stale temporal/TODO residue and house
-style. Violations are reported as `contentPolicy` findings that name the rule broken.
-
-Turn it on with `contentPolicy.enabled: true` in `.postil.yaml` for the built-in
-baseline, or drop repo-specific additions in `.postil/content-policy.md` (which also
-turns it on by itself, the same way `.postil/guardrails.md` does); the file's rules are
-appended to the baseline, not a replacement for it.
+On by default. It reviews human-readable prose in the diff, including Markdown, code
+comments, docstrings, user-facing/log strings, and the PR title/description, never code
+logic or identifiers. It checks for fabricated or contradicted documentation claims,
+self-contradictions the same PR creates, authoring-process narration and AI-authorship
+residue, leaked conversation/transcript text, and (lower severity) stale temporal/TODO
+residue and house style. Violations are reported as `contentPolicy` findings that name
+the rule broken.
+
+The built-in baseline reports fabricated or contradicted claims and conversation leaks
+at `error`, self-contradictions and authorship residue at `warn`, and stale/style residue
+at `info`. With the default `gate.failOn: error`, genuine violations of either error rule
+fail the gate. Repo-specific rules in `.postil/content-policy.md` are appended to the
+baseline, not a replacement for it. Set `contentPolicy.enabled: false` in `.postil.yaml`
+to fully disable both the baseline and repo-specific additions.
 
 ## Configuration
 
@@ -132,9 +135,10 @@ gate:
   failOn: error           # info | warn | error | never
   onError: block          # block (fail closed, default) | advisory (fail open on
                           # provider outage only; unusable model output still blocks)
-contentPolicy:
-  enabled: false          # true reviews prose in the diff for fabricated claims, AI-
-                          # authorship residue, and stale/style residue (see above)
+# Content policy uses the built-in baseline by default and extends it with
+# .postil/content-policy.md. Uncomment to fully opt out:
+# contentPolicy:
+#   enabled: false
 model:
   name: deepseek/deepseek-v4-pro
   cascade: [anthropic/claude-sonnet-4.6]

ROADMAP.md

@@ -13,11 +13,11 @@
 - SARIF 2.1.0 output (`--sarif`) for code-scanning ingestion.
 - Repo guardrails file (`.postil/guardrails.md`) injected into the prompt; violations
   surface as `kind: guardrail` findings that quote the rule.
-- Content policy (`contentPolicy.enabled` or `.postil/content-policy.md`, off by
-  default): reviews prose in the diff for fabricated/contradicted doc claims,
-  same-PR self-contradictions, AI-authorship/process-narration residue, leaked
-  conversation text, and stale temporal/TODO/style residue; surfaces as
-  `kind: contentPolicy`.
+- Content policy (built-in baseline by default, optional
+  `.postil/content-policy.md` additions, `contentPolicy.enabled: false` opt-out):
+  reviews prose in the diff for fabricated/contradicted doc claims, same-PR
+  self-contradictions, AI-authorship/process-narration residue, leaked conversation
+  text, and stale temporal/TODO/style residue; surfaces as `kind: contentPolicy`.
 - `postil plan` deterministic config dry-run; `postil doctor`; pre-push hook.
 - `.coderabbit.yaml` translation for zero-cost migration.
 - Model cascade + concurrent multi-model consensus over any OpenAI-compatible endpoint;

src/config.rs

@@ -93,7 +93,7 @@ pub struct Config {
     /// rules. Findings that violate one are emitted as `kind: guardrail`.
     pub guardrails: Option<String>,
     /// Active content-policy rules (built-in baseline, optionally extended by
-    /// `.postil/content-policy.md`), or `None` when the dimension is off.
+    /// `.postil/content-policy.md`), or `None` after an explicit opt-out.
     /// Findings are emitted as `kind: contentPolicy`. See [`load`](Self::load).
     pub content_policy: Option<String>,
     /// Set by an explicit `contentPolicy: { enabled: false }`, so a repo can
@@ -122,7 +122,7 @@ impl Default for Config {
             api_base: DEFAULT_API_BASE.to_string(),
             consensus: 1,
             guardrails: None,
-            content_policy: None,
+            content_policy: Some(BUILTIN_CONTENT_POLICY.to_string()),
             content_policy_disabled: false,
             source: "defaults".to_string(),
         }
@@ -180,8 +180,8 @@ pub struct ModelSection {
 /// documentation claims, self-contradictions the same PR creates, authoring-
 /// process narration and AI-authorship residue, leaked conversation/transcript
 /// text, and (low severity) stale temporal/TODO residue and house style.
-/// Off by default; enabling it or dropping a `.postil/content-policy.md` file
-/// turns it on. Findings are reported as `kind: contentPolicy`.
+/// On by default; `contentPolicy: { enabled: false }` fully disables it.
+/// Findings are reported as `kind: contentPolicy`.
 #[derive(Debug, Default, Deserialize, Serialize)]
 #[serde(deny_unknown_fields, rename_all = "camelCase")]
 pub struct ContentPolicySection {
@@ -220,9 +220,8 @@ impl Config {
                 cfg.guardrails = Some(trimmed.to_string());
             }
         }
-        // A repo-specific content-policy file turns the dimension on (like
-        // guardrails.md) even without an explicit `contentPolicy.enabled:
-        // true`, unless that was explicitly set to false above.
+        // Repo-specific content-policy rules extend the built-in baseline,
+        // unless the repository explicitly opted out above.
         if !cfg.content_policy_disabled {
             let content_policy_path = root.join(".postil").join("content-policy.md");
             if let Ok(text) = std::fs::read_to_string(&content_policy_path) {
@@ -334,7 +333,10 @@ impl Config {
         }
         if let Some(cp) = f.content_policy {
             match cp.enabled {
-                Some(true) => self.content_policy = Some(BUILTIN_CONTENT_POLICY.to_string()),
+                Some(true) => {
+                    self.content_policy = Some(BUILTIN_CONTENT_POLICY.to_string());
+                    self.content_policy_disabled = false;
+                }
                 Some(false) => {
                     self.content_policy = None;
                     self.content_policy_disabled = true;
@@ -471,12 +473,10 @@ model:
   #                                         # trusted local repo.
   # consensus: 1
 
+# Content policy is on by default. Repo rules in .postil/content-policy.md extend
+# the built-in baseline. Uncomment this explicit opt-out to fully disable it.
 # contentPolicy:
-#   enabled: true           # review prose in the diff (docs, comments, PR body) for
-#                           # fabricated claims, AI-authorship residue, leaked
-#                           # conversation text, and stale/style nits. Off by default;
-#                           # also turns on automatically if .postil/content-policy.md
-#                           # exists. See https://postil.dev/docs/content-policy
+#   enabled: false          # See https://postil.dev/docs/content-policy
 "#;
 
 /// Built-in content-policy baseline, active whenever the dimension is on
@@ -631,10 +631,11 @@ mod tests {
     }
 
     #[test]
-    fn content_policy_off_by_default() {
+    fn content_policy_on_by_default() {
         let dir = tempfile::tempdir().unwrap();
         let c = Config::load(dir.path(), None).unwrap();
-        assert!(c.content_policy.is_none());
+        assert_eq!(c.content_policy.as_deref(), Some(BUILTIN_CONTENT_POLICY));
+        assert!(!c.content_policy_disabled);
     }
 
     #[test]
@@ -650,7 +651,19 @@ mod tests {
     }
 
     #[test]
-    fn content_policy_file_presence_turns_it_on_and_is_appended() {
+    fn builtin_content_policy_severity_profile_matches_default_gate() {
+        assert_eq!(BUILTIN_CONTENT_POLICY.matches("report at error").count(), 2);
+        assert_eq!(BUILTIN_CONTENT_POLICY.matches("report at warn").count(), 2);
+        assert_eq!(BUILTIN_CONTENT_POLICY.matches("report at info").count(), 2);
+
+        let c = Config::default();
+        assert!(c.gate_fail_on.fails(Severity::Error));
+        assert!(!c.gate_fail_on.fails(Severity::Warn));
+        assert!(!c.gate_fail_on.fails(Severity::Info));
+    }
+
+    #[test]
+    fn content_policy_file_is_appended_after_the_builtin_baseline() {
         let dir = tempfile::tempdir().unwrap();
         std::fs::create_dir(dir.path().join(".postil")).unwrap();
         write(
@@ -660,8 +673,10 @@ mod tests {
         );
         let c = Config::load(dir.path(), None).unwrap();
         let policy = c.content_policy.as_deref().unwrap();
-        assert!(policy.contains("Fabricated"));
-        assert!(policy.contains("Never mention unreleased pricing"));
+        assert!(policy.starts_with(BUILTIN_CONTENT_POLICY));
+        assert!(
+            policy.contains("--- REPO-SPECIFIC ADDITIONS ---\nNever mention unreleased pricing.")
+        );
     }
 
     #[test]
@@ -697,5 +712,19 @@ mod tests {
         );
         let c = Config::load(dir.path(), None).unwrap();
         assert!(c.content_policy.is_none());
+        assert!(c.content_policy_disabled);
+    }
+
+    #[test]
+    fn content_policy_explicit_false_disables_the_default_baseline() {
+        let dir = tempfile::tempdir().unwrap();
+        write(
+            dir.path(),
+            ".postil.yaml",
+            "contentPolicy:\n  enabled: false\n",
+        );
+        let c = Config::load(dir.path(), None).unwrap();
+        assert!(c.content_policy.is_none());
+        assert!(c.content_policy_disabled);
     }
 }

src/envelope.rs

@@ -42,8 +42,8 @@ pub enum Kind {
     /// Violates the content policy: a fabricated/contradicted doc claim, a
     /// self-contradiction the same PR creates, authoring-process or AI
     /// narration residue, leaked conversation/transcript text, or (low
-    /// severity, opt-in noise) stale temporal/TODO residue and house style.
-    /// Only emitted when `contentPolicy` is active for the repo.
+    /// severity) stale temporal/TODO residue and house style. Only emitted when
+    /// `contentPolicy` is active for the repo.
     ContentPolicy,
 }
 

src/prompt.rs

@@ -261,7 +261,8 @@ mod tests {
 
     #[test]
     fn system_prompt_omits_content_policy_when_inactive() {
-        let cfg = Config::default();
+        let mut cfg = Config::default();
+        cfg.content_policy = None;
         let p = system_prompt(&cfg);
         assert!(!p.contains("CONTENT POLICY"));
     }

tests/e2e.rs

@@ -917,7 +917,7 @@ fn llm_with_summary(summary: &str, findings: Value) -> Value {
 }
 
 // Shared GitHub remote setup for the content-policy PR-description tests. The PR
-// body carries prose the model flags; content policy is enabled via config.
+// body carries prose the model flags; content policy is active by default.
 async fn content_policy_pr_server(llm: Value) -> MockServer {
     let server = MockServer::start().await;
     Mock::given(method("POST"))
@@ -976,11 +976,6 @@ async fn content_policy_pr_body_finding_survives_grounding() {
     .await;
 
     let dir = tempfile::tempdir().unwrap();
-    std::fs::write(
-        dir.path().join(".postil.yaml"),
-        "contentPolicy:\n  enabled: true\n",
-    )
-    .unwrap();
     let out = postil()
         .current_dir(dir.path())
         .env("POSTIL_API_BASE", server.uri())
@@ -1030,16 +1025,11 @@ async fn content_policy_pr_body_finding_survives_grounding() {
 
 #[tokio::test]
 async fn content_policy_clean_run_does_not_fail_close() {
-    // With content policy on and no violations, the run stays clean: the numbered
+    // With default content policy and no violations, the run stays clean: the numbered
     // PR-description block must not induce a spurious ungrounded/fail-closed run.
     let server = content_policy_pr_server(llm_with_summary("", json!([]))).await;
 
     let dir = tempfile::tempdir().unwrap();
-    std::fs::write(
-        dir.path().join(".postil.yaml"),
-        "contentPolicy:\n  enabled: true\n",
-    )
-    .unwrap();
     let out = postil()
         .current_dir(dir.path())
         .env("POSTIL_API_BASE", server.uri())
@@ -1873,6 +1863,7 @@ fn init_writes_starter_and_config_shows_provenance() {
     let stdout = String::from_utf8(out.get_output().stdout.clone()).unwrap();
     assert!(stdout.contains("source: .postil.yaml"));
     assert!(stdout.contains("gate.failOn: error"));
+    assert!(stdout.contains("contentPolicy: active"));
 }
 
 #[test]