feat(review): make fast mode best effort

README.md

@@ -546,6 +546,7 @@ Run `cr config show` to inspect the active profile and credential status.
 ```yaml
 profiles:
   default:
+    fast: false
     git:
       host: github.com
       auth_mode: pat
@@ -1111,7 +1112,8 @@ Modes:
 | `--rerun` | Bypass existing local approval, approval-override, resume, and marker gates and start a new live review while retaining provider-session reuse. Mutually exclusive with `--retry-posts`. |
 | `--retry-posts` | Retry missing or failed required posts for an existing run without rerunning LLM planning or checking approval overrides. Mutually exclusive with `--rerun` and incompatible with `--session`. |
 | `--fresh-session` | Start a fresh provider conversation for this invocation without changing local review gates. Incompatible with `--retry-posts`, which does not run LLM planning. |
-| `--fast` | Request fast execution for reviewer agents. Incompatible with `--retry-posts`; unsupported runtime/model combinations fail before LLM planning. |
+| `--fast` | Enable fast execution for reviewer agents, overriding the profile default. Incompatible with `--retry-posts`. |
+| `--no-fast` | Disable fast execution for reviewer agents, overriding the profile default. Mutually exclusive with `--fast`. |
 
 Review selection and execution flags:
 
@@ -1144,18 +1146,24 @@ Policy and output flags:
 | `--no-resolve-threads` | Do not plan thread-resolution actions. Also implied by profile `resolve_threads: never`. |
 | `--json` | Emit JSON. |
 
-Fast mode currently supports `claude_cli` and `anthropic_api` with
-`claude-opus-4-8` or `claude-opus-4-7`; Anthropic has deprecated Opus 4.7 fast
-mode and plans to remove it on July 24, 2026. Claude CLI receives a per-session
-`fastMode` setting, while Anthropic API requests use its fast-mode beta. Fast
-mode has premium pricing and applies only to reviewer agents, not selection,
-synthesis, approval-override classification, or `cr respond` thread analysis.
+Fast mode defaults off; set `fast: true` on a profile to enable it by default.
+`--fast` and `--no-fast` override the profile. Unsupported runtime/model
+combinations warn and continue at normal speed. Fast mode supports `claude_cli`
+and `anthropic_api` with `claude-opus-4-8` or `claude-opus-4-7`, and `codex_cli`
+with `gpt-5.4`, `gpt-5.5`, `gpt-5.6-sol`, `gpt-5.6-terra`, or `gpt-5.6-luna`.
+Anthropic has deprecated Opus 4.7 fast mode and plans to remove it on July 24,
+2026. Claude CLI receives a per-session `fastMode` setting, Anthropic API
+requests use its fast-mode beta, and Codex CLI receives `service_tier="fast"`.
+Fast mode has premium pricing and applies only to reviewer agents, not
+selection, synthesis, approval-override classification, or `cr respond` thread
+analysis.
 Anthropic API fast mode is a research preview that requires account access;
 requests without access fail as upstream errors during the run. Subscription
 fast mode bills usage credits, requires owner enablement on Team and Enterprise,
 and can silently degrade to standard speed when credits or fast capacity are
-unavailable. Reviewer runtime artifacts therefore record both the fast request
-and the speed actually reported by the provider, or `unknown` when unavailable.
+unavailable. Reviewer runtime artifacts therefore record the fast request,
+whether it was ignored as unsupported, and the speed actually reported by the
+provider, or `unknown` when unavailable.
 
 Local run state and provider session state are independent. By default, each
 PR/profile/posting-identity tuple gets one durable provider session shared by

docs/init-config-surface.md

@@ -57,6 +57,7 @@ intentionally unsupported.
 | config.profiles.<name>.reviewer.github_app_installation.mode | Review profile composition chooses how a GitHub App reviewer entity resolves its installation. | Interactive review-profile flow only for now. | Required when `reviewer.entity` uses GitHub App auth. New GitHub App reviewer selections default to `discover_from_repository`. | `discover_from_repository` resolves from PR repository context; `pinned` requires `installation_id`. Must be empty for PAT and Git identity reviewers. | GitHub App reviewer installation routing tests. |
 | config.profiles.<name>.reviewer.github_app_installation.installation_id | Review profile composition stores an optional pinned GitHub App installation id for this profile. | Interactive review-profile flow only for now. | Empty unless installation mode is `pinned`. | Must be a decimal GitHub App installation id when present. This is profile routing data, not secret material. | GitHub App reviewer installation routing tests. |
 | config.profiles.<name>.llm_runtime | Review profile composition selects one configured LLM runtime by name. | Interactive review-profile flow only for now. Future config commands may own scripted profile composition. | New profiles require an explicit runtime choice. Existing runtime reference is pre-populated. | Must reference `config.llm_runtimes`. API-key runtime credentials must differ from Git and selected reviewer entity credentials when the store also matches. | First-class LLM-runtime reference validation tests. |
+| config.profiles.<name>.fast | Not shown in interactive init; fast mode is a power-user profile default. | Direct config-file management; `cr review --fast` and `--no-fast` override it per invocation. | Omitted defaults to `false`. | Unsupported runtime/model combinations warn and continue at normal speed. | Profile config and review command precedence tests. |
 | config.profiles.<name>.agent_sources[] | Direct trusted-directory editor shows existing profile agent-source paths in one multiline field with explanatory notes about trust and separate repo-local `.codereview/agents` discovery. | Existing `cr init --agent-source`; existing `cr config agent-source add/remove`; #186 docs sequence. | Omitted means no additional profile-specific trusted directories. Existing values are pre-populated line-by-line. | Preserve on skip. Clearing the field removes all profile-specific agent sources. Entered paths are normalized and deduped through the shared config-edit helper. | #177 extracts helper. #183 tests add/remove/reset/preserve. #289 prompt tests cover explanatory copy, multiline entry, and Back. |
 | config.profiles.<name>.hooks[].event | Not shown in interactive init; lifecycle hooks are power-user config. | Direct config-file management. | No hooks by default. | Must be one of the documented review or respond lifecycle events; the benchmark namespace is reserved. | Hook validation and runtime dispatch tests. |
 | config.profiles.<name>.hooks[].argv[] | Not shown in interactive init. | Direct config-file management. | Required for each hook. | Executed directly without a shell; the command must be non-empty and no argument may contain NUL. | Hook validation and argv execution tests. |

internal/app/runtime_test.go

@@ -949,6 +949,7 @@ func TestAdapterConstructorsCoverLLMRuntimeSpecs(t *testing.T) {
 	wantFastModels := map[config.LLMAdapter][]string{
 		config.LLMAdapterClaudeCLI:    {"claude-opus-4-8", "claude-opus-4-7"},
 		config.LLMAdapterAnthropicAPI: {"claude-opus-4-8", "claude-opus-4-7"},
+		config.LLMAdapterCodexCLI:     {"gpt-5.4", "gpt-5.5", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"},
 	}
 	for _, spec := range config.LLMRuntimeSpecs() {
 		if _, duplicate := want[spec.Adapter]; duplicate {

internal/cmd/reviewcmd/reviewcmd.go

@@ -43,8 +43,9 @@ including after pushes and across dry-run/live invocations. --session selects
 a named live-review session instead; --fresh-session starts a fresh provider
 conversation for this invocation.
 
---fast requests fast execution for reviewer agents only. The configured
-adapter and every possible reviewer model must support it.
+--fast requests fast execution for reviewer agents only; --no-fast disables it.
+CLI flags override the profile default. Unsupported runtimes or reviewer models
+emit a warning and continue at normal speed.
 
 If no existing approval is present and a prior codereview marker exists from
 the posting identity, cr can fast-path approval when the PR author posts an
@@ -66,6 +67,7 @@ type commandFlags struct {
 	retryPosts        bool
 	freshSession      bool
 	fast              bool
+	noFast            bool
 	agentsDirs        []string
 	failOn            string
 	sessionName       string
@@ -108,6 +110,7 @@ func RegisterWithFactory(rootCmd *cobra.Command, opts *root.Options, factory Run
 	cmd.Flags().BoolVar(&flags.retryPosts, "retry-posts", false, "Retry missing or failed required posts without rerunning review or checking approval overrides")
 	cmd.Flags().BoolVar(&flags.freshSession, "fresh-session", false, "Start a fresh provider conversation without changing local review gates")
 	cmd.Flags().BoolVar(&flags.fast, "fast", false, "Request fast execution for supported reviewer runtimes and models")
+	cmd.Flags().BoolVar(&flags.noFast, "no-fast", false, "Disable fast execution for reviewer agents")
 	cmd.Flags().StringArrayVar(&flags.agentsDirs, "agents-dir", nil, "Additional trusted agents directory")
 	cmd.Flags().StringVar(&flags.failOn, "fail-on", "", "Exit 1 when a finding at or above severity exists")
 	cmd.Flags().StringVar(&flags.sessionName, "session", "", "Override the PR's default LLM session with a named live-review session")
@@ -129,6 +132,9 @@ func RegisterWithFactory(rootCmd *cobra.Command, opts *root.Options, factory Run
 }
 
 func runReview(ctx context.Context, cmd *cobra.Command, opts *root.Options, factory RuntimeFactory, flags commandFlags, prArg string) error {
+	if cmd.Flags().Changed("fast") && cmd.Flags().Changed("no-fast") {
+		return exitcode.Usage(fmt.Errorf("--fast and --no-fast are mutually exclusive"))
+	}
 	if flags.noPost {
 		flags.dryRun = true
 	}
@@ -207,9 +213,6 @@ func runReview(ctx context.Context, cmd *cobra.Command, opts *root.Options, fact
 	if flags.freshSession && flags.retryPosts {
 		return exitcode.Usage(fmt.Errorf("--fresh-session cannot be used with --retry-posts"))
 	}
-	if flags.fast && flags.retryPosts {
-		return exitcode.Usage(fmt.Errorf("--fast cannot be used with --retry-posts"))
-	}
 	if flags.maxAgents < 0 {
 		return exitcode.Usage(fmt.Errorf("--max-agents must be non-negative"))
 	}
@@ -266,6 +269,15 @@ func runReview(ctx context.Context, cmd *cobra.Command, opts *root.Options, fact
 		return exitcode.Usage(profileSpan.End(err))
 	}
 	_ = profileSpan.End(nil)
+	reviewerFast := profile.Fast
+	if cmd.Flags().Changed("fast") {
+		reviewerFast = flags.fast
+	} else if cmd.Flags().Changed("no-fast") {
+		reviewerFast = !flags.noFast
+	}
+	if reviewerFast && flags.retryPosts {
+		return exitcode.Usage(fmt.Errorf("fast mode cannot be used with --retry-posts"))
+	}
 
 	runtimeReq := app.OpenRequest{
 		Config:                            cfg,
@@ -316,7 +328,7 @@ func runReview(ctx context.Context, cmd *cobra.Command, opts *root.Options, fact
 		ReviewerModelOverride:       reviewerModel,
 		ReviewerModelTierOverride:   reviewerModelTier,
 		ReviewerEffortOverride:      reviewerEffort,
-		ReviewerFast:                flags.fast,
+		ReviewerFast:                reviewerFast,
 		ReviewBaseSHA:               reviewBaseSHA,
 		ReviewHeadSHA:               reviewHeadSHA,
 		Rerun:                       flags.rerun,

internal/cmd/reviewcmd/reviewcmd_test.go

@@ -926,6 +926,69 @@ func TestReviewFastPropagatesToPipelineRequest(t *testing.T) {
 	}
 }
 
+func TestReviewFastPreferencePrecedence(t *testing.T) {
+	for _, tt := range []struct {
+		name        string
+		profileFast bool
+		flags       []string
+		want        bool
+	}{
+		{name: "default off"},
+		{name: "profile default", profileFast: true, want: true},
+		{name: "flag enables", flags: []string{"--fast"}, want: true},
+		{name: "explicit fast false overrides profile", profileFast: true, flags: []string{"--fast=false"}},
+		{name: "no-fast overrides profile", profileFast: true, flags: []string{"--no-fast"}},
+		{name: "explicit no-fast false enables", flags: []string{"--no-fast=false"}, want: true},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := testConfig()
+			profile := cfg.Profiles["home"]
+			profile.Fast = tt.profileFast
+			cfg.Profiles["home"] = profile
+			runner := &fakeRunner{result: testPipelineResult(false)}
+			cmd, _ := newTestCommand(t, cfg, fakeFactory(runner))
+			args := []string{"review", "https://github.com/open-cli-collective/codereview-cli/pull/29", "--dry-run"}
+			args = append(args, tt.flags...)
+			if err := root.Execute(cmd, args); err != nil {
+				t.Fatalf("Execute: %v", err)
+			}
+			if len(runner.requests) != 1 || runner.requests[0].ReviewerFast != tt.want {
+				t.Fatalf("requests = %#v, want ReviewerFast=%t", runner.requests, tt.want)
+			}
+		})
+	}
+}
+
+func TestReviewRetryPostsUsesEffectiveFastPreference(t *testing.T) {
+	cfg := testConfig()
+	profile := cfg.Profiles["home"]
+	profile.Fast = true
+	cfg.Profiles["home"] = profile
+
+	t.Run("profile fast rejects before runtime construction", func(t *testing.T) {
+		factoryCalled := false
+		cmd, _ := newTestCommand(t, cfg, func(context.Context, app.OpenRequest) (app.Runtime, error) {
+			factoryCalled = true
+			return app.Runtime{}, nil
+		})
+		err := root.Execute(cmd, []string{"review", "https://github.com/open-cli-collective/codereview-cli/pull/29", "--retry-posts"})
+		if exitcode.FromError(err) != exitcode.UsageError || factoryCalled {
+			t.Fatalf("Execute error = %v, factory called = %t, want usage before runtime", err, factoryCalled)
+		}
+	})
+
+	t.Run("no-fast permits retry", func(t *testing.T) {
+		runner := &fakeRunner{liveResult: testLiveResult(false)}
+		cmd, _ := newTestCommand(t, cfg, fakeFactory(runner))
+		if err := root.Execute(cmd, []string{"review", "https://github.com/open-cli-collective/codereview-cli/pull/29", "--retry-posts", "--no-fast"}); err != nil {
+			t.Fatalf("Execute: %v", err)
+		}
+		if len(runner.liveRequests) != 1 || runner.liveRequests[0].ReviewerFast {
+			t.Fatalf("live requests = %#v, want standard-speed retry", runner.liveRequests)
+		}
+	})
+}
+
 func TestReviewLiveSessionPassesNamedSession(t *testing.T) {
 	runner := &fakeRunner{liveResult: testLiveResult(false)}
 	cmd, _ := newTestCommand(t, testConfig(), fakeFactory(runner))
@@ -973,6 +1036,7 @@ func TestReviewRejectsInvalidInputs(t *testing.T) {
 		{name: "session retry posts", args: []string{"review", "https://github.com/open-cli-collective/codereview-cli/pull/29", "--retry-posts", "--session", "daily"}},
 		{name: "fresh session retry posts", args: []string{"review", "https://github.com/open-cli-collective/codereview-cli/pull/29", "--retry-posts", "--fresh-session"}},
 		{name: "fast retry posts", args: []string{"review", "https://github.com/open-cli-collective/codereview-cli/pull/29", "--retry-posts", "--fast"}},
+		{name: "fast flag conflict", args: []string{"review", "https://github.com/open-cli-collective/codereview-cli/pull/29", "--dry-run", "--fast", "--no-fast"}},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {

internal/config/config.go

@@ -146,6 +146,7 @@ type Profile struct {
 	Git              GitConfig       `yaml:"-" json:"-"`
 	Reviewer         ProfileReviewer `yaml:"reviewer" json:"reviewer"`
 	LLMRuntime       string          `yaml:"llm_runtime" json:"llm_runtime"`
+	Fast             bool            `yaml:"fast,omitempty" json:"fast,omitempty"`
 	AgentSources     []string        `yaml:"agent_sources,omitempty" json:"agent_sources,omitempty"`
 	ReviewPolicy     ReviewPolicy    `yaml:"review_policy,omitempty" json:"review_policy"`
 	Hooks            []Hook          `yaml:"hooks,omitempty" json:"hooks,omitempty"`
@@ -163,6 +164,7 @@ type profileYAML struct {
 	Git              GitConfig       `yaml:"git,omitempty" json:"git,omitempty"`
 	Reviewer         ProfileReviewer `yaml:"reviewer" json:"reviewer"`
 	LLMRuntime       string          `yaml:"llm_runtime" json:"llm_runtime"`
+	Fast             bool            `yaml:"fast,omitempty" json:"fast,omitempty"`
 	AgentSources     []string        `yaml:"agent_sources,omitempty" json:"agent_sources,omitempty"`
 	ReviewPolicy     ReviewPolicy    `yaml:"review_policy,omitempty" json:"review_policy"`
 	Hooks            []Hook          `yaml:"hooks,omitempty" json:"hooks,omitempty"`
@@ -210,6 +212,7 @@ func (p *Profile) UnmarshalYAML(value *yaml.Node) error {
 		Git:              raw.Git,
 		Reviewer:         raw.Reviewer,
 		LLMRuntime:       raw.LLMRuntime,
+		Fast:             raw.Fast,
 		AgentSources:     raw.AgentSources,
 		ReviewPolicy:     raw.ReviewPolicy,
 		Hooks:            raw.Hooks,
@@ -224,6 +227,7 @@ func (p Profile) MarshalYAML() (any, error) {
 		RepositoryAccess: p.RepositoryAccess,
 		Reviewer:         p.Reviewer,
 		LLMRuntime:       p.LLMRuntime,
+		Fast:             p.Fast,
 		AgentSources:     p.AgentSources,
 		ReviewPolicy:     p.ReviewPolicy,
 		Hooks:            p.Hooks,
@@ -561,6 +565,7 @@ var llmRuntimeSpecs = []LLMRuntimeSpec{
 			string(ModelTierMedium): "gpt-5.4",
 			string(ModelTierLarge):  "gpt-5.5",
 		},
+		FastModeModels: []string{"gpt-5.4", "gpt-5.5", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"},
 	},
 	{
 		Provider:      LLMProviderOpenAI,

internal/config/config_test.go

@@ -217,6 +217,7 @@ func TestLoadAcceptsCodexCLISubscriptionProfile(t *testing.T) {
     adapter: codex_cli
 profiles:
   codex:
+    fast: true
     git:
       host: github.com
       auth_mode: pat
@@ -232,6 +233,20 @@ profiles:
 		t.Fatalf("Load: %v", err)
 	}
 	profile := cfg.Profiles["codex"]
+	if !profile.Fast {
+		t.Fatal("Fast = false, want true")
+	}
+	savedPath := filepath.Join(t.TempDir(), "saved.yml")
+	if err := Save(savedPath, cfg); err != nil {
+		t.Fatalf("Save: %v", err)
+	}
+	saved, err := Load(savedPath)
+	if err != nil {
+		t.Fatalf("reload saved config: %v", err)
+	}
+	if !saved.Profiles["codex"].Fast {
+		t.Fatal("saved Fast = false, want true")
+	}
 	if profile.LLM.Provider != LLMProviderOpenAI || profile.LLM.Adapter != LLMAdapterCodexCLI {
 		t.Fatalf("LLM = %#v, want openai/codex_cli", profile.LLM)
 	}

internal/llmadapters/subprocess.go

@@ -659,6 +659,9 @@ func (a *SubprocessAdapter) buildArgsForSession(req Request, scratch string, res
 		if req.Effort != "" {
 			args = append(args, "-c", "model_reasoning_effort="+req.Effort)
 		}
+		if req.Fast {
+			args = append(args, "-c", `service_tier="fast"`)
+		}
 		if resumeSessionID != "" {
 			args = subprocessCodexResumeArgs()
 			if req.Model != "" {
@@ -667,6 +670,9 @@ func (a *SubprocessAdapter) buildArgsForSession(req Request, scratch string, res
 			if req.Effort != "" {
 				args = append(args, "-c", "model_reasoning_effort="+req.Effort)
 			}
+			if req.Fast {
+				args = append(args, "-c", `service_tier="fast"`)
+			}
 			args = append(args, resumeSessionID)
 		}
 		return append(args, "--", req.Prompt), nil