feat(ambient): pipeline flow, SessionStart classification, minimal preamble

CLAUDE.md

@@ -40,6 +40,8 @@ Commands with Teams Variant ship as `{name}.md` (parallel subagents) and `{name}
 
 **Working Memory**: Three shell-script hooks (`scripts/hooks/`) provide automatic session continuity. Toggleable via `devflow memory --enable/--disable/--status` or `devflow init --memory/--no-memory`. Stop hook → reads last turn from session transcript (`~/.claude/projects/{encoded-cwd}/{session_id}.jsonl`), spawns background `claude -p --model haiku` to update `.memory/WORKING-MEMORY.md` with structured sections (`## Now`, `## Progress`, `## Decisions`, `## Modified Files`, `## Context`, `## Session Log`; throttled: skips if triggered <2min ago; concurrent sessions serialize via mkdir-based lock). SessionStart hook → injects previous memory + git state as `additionalContext` on `/clear`, startup, or compact (warns if >1h stale; injects pre-compact memory snapshot when compaction happened mid-session). PreCompact hook → saves git state + WORKING-MEMORY.md snapshot + bootstraps minimal WORKING-MEMORY.md if none exists. Zero-ceremony context preservation.
 
+**Ambient Mode**: Three-layer architecture for always-on intent classification. SessionStart hook (`session-start-classification`) reads lean classification rules (`~/.claude/skills/devflow:router/references/classification-rules.md`, ~30 lines) and injects as `additionalContext` — once per session, deterministic, zero model overhead. UserPromptSubmit hook (`preamble`) injects a one-sentence prompt per message triggering classification + router loading via Skill tool. Router SKILL.md is a pure skill lookup table (~50 lines) loaded on-demand only for GUIDED/ORCHESTRATED depth — maps intent×depth to domain and orchestration skills. Toggleable via `devflow ambient --enable/--disable/--status` or `devflow init`.
+
 **Self-Learning**: A SessionEnd hook (`session-end-learning`) accumulates session IDs and triggers a background `claude -p --model sonnet` every 3 sessions (5 at 15+ observations) to detect repeated workflows and procedural knowledge from batch transcripts. Observations accumulate in `.memory/learning-log.jsonl` with confidence scores, temporal decay, and daily run caps. When confidence thresholds are met (5 observations with 7-day temporal spread for both workflow and procedural types), artifacts are auto-created as slash commands (`.claude/commands/self-learning/`) or skills (`.claude/skills/{slug}/`). Loaded artifacts are reinforced locally (no LLM) on each session end. Single toggle mechanism: hook presence in `settings.json` IS the enabled state — no `enabled` field in `learning.json`. Toggleable via `devflow learn --enable/--disable/--status` or `devflow init --learn/--no-learn`. Configurable model/throttle/caps/debug via `devflow learn --configure`. Use `devflow learn --reset` to remove all artifacts + log + transient state. Use `devflow learn --purge` to remove invalid observations. Debug logs stored at `~/.devflow/logs/{project-slug}/`.
 
 **Claude Code Flags**: Typed registry (`src/cli/utils/flags.ts`) for managing Claude Code feature flags (env vars and top-level settings). Pure functions `applyFlags`/`stripFlags`/`getDefaultFlags` follow the `applyTeamsConfig`/`stripTeamsConfig` pattern. Initial flags: `tool-search`, `lsp`, `clear-context-on-plan` (default ON), `brief`, `disable-1m-context` (default OFF). Manageable via `devflow flags --enable/--disable/--status/--list`. Stored in manifest `features.flags: string[]`.
@@ -55,7 +57,7 @@ devflow/
 ├── plugins/devflow-*/      # 17 plugins (8 core + 9 optional language/ecosystem)
 ├── docs/reference/         # Detailed reference documentation
 ├── scripts/                # Helper scripts (statusline, docs-helpers)
-│   └── hooks/              # Working Memory + ambient + learning hooks (stop, session-start, pre-compact, preamble, session-end-learning, stop-update-learning [deprecated], background-learning)
+│   └── hooks/              # Working Memory + ambient + learning hooks (stop, session-start-memory, session-start-classification, pre-compact, preamble, session-end-learning, stop-update-learning [deprecated], background-learning)
 ├── src/cli/                # TypeScript CLI (init, list, uninstall, ambient, learn, flags)
 ├── .claude-plugin/         # Marketplace registry
 ├── .docs/                  # Project docs (reviews, design) — per-project

plugins/devflow-ambient/README.md

@@ -1,11 +1,11 @@
 # devflow-ambient
 
-Ambient mode — classifies intent and applies proportional effort via a `UserPromptSubmit` hook. No slash command — ambient mode activates automatically on every prompt when enabled.
+Ambient mode — classifies intent and applies proportional effort via a `SessionStart` hook and a `UserPromptSubmit` hook. No slash command — ambient mode activates automatically on every prompt when enabled.
 
 ## Activation
 
 ```bash
-devflow ambient --enable    # Register UserPromptSubmit hook
+devflow ambient --enable    # Register ambient mode hooks
 devflow ambient --disable   # Remove hook
 devflow ambient --status    # Check if enabled
 ```
@@ -44,9 +44,9 @@ Skills are loaded via the Skill tool and work happens in the main session:
 
 | Intent | Skills | Main Session Work | Post-Work |
 |--------|--------|-------------------|-----------|
-| IMPLEMENT | test-driven-development, patterns, research | Implement with TDD | `Task(subagent_type="Simplifier")` |
-| DEBUG | software-design, testing | Investigate, diagnose, fix | `Task(subagent_type="Simplifier")` |
-| PLAN | patterns, software-design | Explore and design | — |
+| IMPLEMENT | test-driven-development, patterns, research | Implement with TDD | `Agent(subagent_type="Simplifier")` |
+| DEBUG | test-driven-development, software-design, testing | Investigate, diagnose, fix | `Agent(subagent_type="Simplifier")` |
+| PLAN | test-driven-development, patterns, software-design, security | Explore and design | — |
 | REVIEW | quality-gates, software-design | Review directly | — |
 
 ## ORCHESTRATED Pipelines

plugins/devflow-code-review/commands/code-review-teams.md

@@ -34,7 +34,7 @@ Run a comprehensive code review of the current branch by spawning a review team
 For each reviewable worktree, spawn Git agent:
 
 ```
-Task(subagent_type="Git", run_in_background=false):
+Agent(subagent_type="Git", run_in_background=false):
 "OPERATION: ensure-pr-ready
 WORKTREE_PATH: {worktree_path}  (omit if cwd)
 Validate branch, commit if needed, push, create PR if needed.
@@ -218,7 +218,7 @@ Spawn 2 agents **in a single message**:
 
 **Git Agent (PR Comments)**:
 ```
-Task(subagent_type="Git", run_in_background=false):
+Agent(subagent_type="Git", run_in_background=false):
 "OPERATION: comment-pr
 WORKTREE_PATH: {worktree_path}  (omit if cwd)
 Read reviews from {worktree_path}/.docs/reviews/{branch_slug}/{timestamp}/

plugins/devflow-code-review/commands/code-review.md

@@ -34,7 +34,7 @@ Run a comprehensive code review of the current branch by spawning parallel revie
 For each reviewable worktree, spawn Git agent:
 
 ```
-Task(subagent_type="Git", run_in_background=false):
+Agent(subagent_type="Git", run_in_background=false):
 "OPERATION: ensure-pr-ready
 WORKTREE_PATH: {worktree_path}  (omit if cwd)
 Validate branch, commit if needed, push, create PR if needed.
@@ -109,7 +109,7 @@ Spawn Reviewer agents **in a single message**. Always run 7 core reviews; condit
 
 Each Reviewer invocation (all in one message, **NOT background**):
 ```
-Task(subagent_type="Reviewer", run_in_background=false):
+Agent(subagent_type="Reviewer", run_in_background=false):
 "Review focusing on {focus}. Load the pattern skill for your focus from the Focus Areas table.
 Follow 6-step process from devflow:review-methodology.
 PR: #{pr_number}, Base: {base_branch}
@@ -126,7 +126,7 @@ In multi-worktree mode, process worktrees **sequentially** (one worktree at a ti
 
 **Git Agent (PR Comments)** per worktree:
 ```
-Task(subagent_type="Git", run_in_background=false):
+Agent(subagent_type="Git", run_in_background=false):
 "OPERATION: comment-pr
 WORKTREE_PATH: {worktree_path}  (omit if cwd)
 Read reviews from {worktree_path}/.docs/reviews/{branch-slug}/{timestamp}/
@@ -139,7 +139,7 @@ Check for existing inline comments at same file:line before creating new ones to
 
 **Synthesizer Agent** per worktree:
 ```
-Task(subagent_type="Synthesizer", run_in_background=false):
+Agent(subagent_type="Synthesizer", run_in_background=false):
 "Mode: review
 WORKTREE_PATH: {worktree_path}  (omit if cwd)
 REVIEW_BASE_DIR: {worktree_path}/.docs/reviews/{branch-slug}/{timestamp}

plugins/devflow-debug/commands/debug-teams.md

@@ -32,7 +32,7 @@ Read `.memory/knowledge/decisions.md` and `.memory/knowledge/pitfalls.md`. Known
 If `$ARGUMENTS` starts with `#`, fetch the GitHub issue:
 
 ```
-Task(subagent_type="Git"):
+Agent(subagent_type="Git"):
 "OPERATION: fetch-issue
 ISSUE: {issue number}
 Return issue title, body, labels, and any linked error logs."

plugins/devflow-debug/commands/debug.md

@@ -32,7 +32,7 @@ Read `.memory/knowledge/decisions.md` and `.memory/knowledge/pitfalls.md`. Known
 If `$ARGUMENTS` starts with `#`, fetch the GitHub issue:
 
 ```
-Task(subagent_type="Git"):
+Agent(subagent_type="Git"):
 "OPERATION: fetch-issue
 ISSUE: {issue number}
 Return issue title, body, labels, and any linked error logs."
@@ -48,7 +48,7 @@ Analyze the bug description (from arguments or issue) and identify 3-5 plausible
 Spawn one Explore agent per hypothesis in a **single message** (parallel execution):
 
 ```
-Task(subagent_type="Explore"):
+Agent(subagent_type="Explore"):
 "Investigate this bug: {bug_description}
 
 Hypothesis: {hypothesis A description}
@@ -67,15 +67,15 @@ Return a structured report:
 - Evidence AGAINST: [list with file:line refs]
 - Key finding: {one-sentence summary}"
 
-Task(subagent_type="Explore"):
+Agent(subagent_type="Explore"):
 "Investigate this bug: {bug_description}
 
 Hypothesis: {hypothesis B description}
 Focus area: {specific code area, mechanism, or condition}
 
 [same steps and return format]"
 
-Task(subagent_type="Explore"):
+Agent(subagent_type="Explore"):
 "Investigate this bug: {bug_description}
 
 Hypothesis: {hypothesis C description}
@@ -91,7 +91,7 @@ Focus area: {specific code area, mechanism, or condition}
 Once all investigators return, spawn a Synthesizer agent to aggregate findings:
 
 ```
-Task(subagent_type="Synthesizer"):
+Agent(subagent_type="Synthesizer"):
 "You are a root cause analyst. Synthesize these investigation reports:
 
 {paste all investigator reports}

plugins/devflow-implement/commands/implement-teams.md

@@ -30,7 +30,7 @@ Record the current branch name as `BASE_BRANCH` - this will be the PR target.
 Spawn Git agent to set up task environment. The Git agent derives the branch name automatically from the issue or task description:
 
 ```
-Task(subagent_type="Git"):
+Agent(subagent_type="Git"):
 "OPERATION: setup-task
 BASE_BRANCH: {current branch name}
 ISSUE_INPUT: {issue number if $ARGUMENTS starts with #, otherwise omit}
@@ -51,7 +51,7 @@ Return the branch setup summary."
 Spawn Skimmer agent for codebase overview:
 
 ```
-Task(subagent_type="Skimmer"):
+Agent(subagent_type="Skimmer"):
 "Orient in codebase for: {task description}
 Run rskim on source directories (NOT repo root) to identify relevant files, functions, integration points"
 ```
@@ -156,7 +156,7 @@ Step 3: GATE — Verify TeamDelete succeeded
 You MUST spawn the Synthesizer agent.
 
 ```
-Task(subagent_type="Synthesizer"):
+Agent(subagent_type="Synthesizer"):
 "Synthesize EXPLORATION outputs for: {task}
 Mode: exploration
 Explorer consensus: {team exploration consensus output}
@@ -252,7 +252,7 @@ Step 3: GATE — Verify TeamDelete succeeded
 You MUST spawn the Synthesizer agent.
 
 ```
-Task(subagent_type="Synthesizer"):
+Agent(subagent_type="Synthesizer"):
 "Synthesize PLANNING outputs for: {task}
 Mode: planning
 Planner consensus: {team planning consensus output}
@@ -282,7 +282,7 @@ Based on Phase 6 synthesis, use the three-strategy framework:
 **SINGLE_CODER** (default):
 
 ```
-Task(subagent_type="Coder"):
+Agent(subagent_type="Coder"):
 "TASK_ID: {task-id}
 TASK_DESCRIPTION: {description}
 BASE_BRANCH: {base branch}
@@ -300,7 +300,7 @@ Spawn Coders one at a time, passing handoff summaries between phases:
 
 **Phase 1 Coder:**
 ```
-Task(subagent_type="Coder"):
+Agent(subagent_type="Coder"):
 "TASK_ID: {task-id}
 TASK_DESCRIPTION: {phase 1 description}
 BASE_BRANCH: {base branch}
@@ -313,7 +313,7 @@ HANDOFF_REQUIRED: true"
 
 **Phase 2+ Coders** (after prior phase completes):
 ```
-Task(subagent_type="Coder"):
+Agent(subagent_type="Coder"):
 "TASK_ID: {task-id}
 TASK_DESCRIPTION: {phase N description}
 BASE_BRANCH: {base branch}
@@ -335,7 +335,7 @@ HANDOFF_REQUIRED: {true if not last phase}"
 Spawn multiple Coders **in a single message**, each with independent subtask:
 
 ```
-Task(subagent_type="Coder"):  # Coder 1
+Agent(subagent_type="Coder"):  # Coder 1
 "TASK_ID: {task-id}-part1
 TASK_DESCRIPTION: {independent subtask 1}
 BASE_BRANCH: {base branch}
@@ -344,7 +344,7 @@ PATTERNS: {patterns}
 CREATE_PR: false
 DOMAIN: {subtask 1 domain}"
 
-Task(subagent_type="Coder"):  # Coder 2 (same message)
+Agent(subagent_type="Coder"):  # Coder 2 (same message)
 "TASK_ID: {task-id}-part2
 TASK_DESCRIPTION: {independent subtask 2}
 BASE_BRANCH: {base branch}
@@ -365,7 +365,7 @@ DOMAIN: {subtask 2 domain}"
 After Coder completes, spawn Validator to verify correctness:
 
 ```
-Task(subagent_type="Validator", model="haiku"):
+Agent(subagent_type="Validator", model="haiku"):
 "FILES_CHANGED: {list of files from Coder output}
 VALIDATION_SCOPE: full
 Run build, typecheck, lint, test. Report pass/fail with failure details."
@@ -377,7 +377,7 @@ Run build, typecheck, lint, test. Report pass/fail with failure details."
 3. If `validation_retry_count <= 2`:
    - Spawn Coder with fix context:
    ```
-   Task(subagent_type="Coder"):
+   Agent(subagent_type="Coder"):
    "TASK_ID: {task-id}
    TASK_DESCRIPTION: Fix validation failures
    OPERATION: validation-fix
@@ -395,7 +395,7 @@ Run build, typecheck, lint, test. Report pass/fail with failure details."
 After validation passes, spawn Simplifier to polish the code:
 
 ```
-Task(subagent_type="Simplifier"):
+Agent(subagent_type="Simplifier"):
 "Simplify recently implemented code
 Task: {task description}
 FILES_CHANGED: {list of files from Coder output}
@@ -407,7 +407,7 @@ Focus on code modified by Coder, apply project standards, enhance clarity"
 After Simplifier completes, spawn Scrutinizer as final quality gate:
 
 ```
-Task(subagent_type="Scrutinizer"):
+Agent(subagent_type="Scrutinizer"):
 "TASK_DESCRIPTION: {task description}
 FILES_CHANGED: {list of files from Coder output}
 Evaluate 9 pillars, fix P0/P1 issues, report status"
@@ -420,7 +420,7 @@ If Scrutinizer returns BLOCKED, report to user and halt.
 If Scrutinizer made code changes (status: FIXED), spawn Validator to verify:
 
 ```
-Task(subagent_type="Validator", model="haiku"):
+Agent(subagent_type="Validator", model="haiku"):
 "FILES_CHANGED: {files modified by Scrutinizer}
 VALIDATION_SCOPE: changed-only
 Verify Scrutinizer's fixes didn't break anything."
@@ -503,7 +503,7 @@ Step 3: GATE — Verify TeamDelete succeeded
 3. If `alignment_fix_count <= 2`:
    - Spawn Coder to fix misalignments:
    ```
-   Task(subagent_type="Coder"):
+   Agent(subagent_type="Coder"):
    "TASK_ID: {task-id}
    TASK_DESCRIPTION: Fix alignment issues
    OPERATION: alignment-fix
@@ -513,7 +513,7 @@ Step 3: GATE — Verify TeamDelete succeeded
    ```
    - Spawn Validator to verify fix didn't break tests:
    ```
-   Task(subagent_type="Validator", model="haiku"):
+   Agent(subagent_type="Validator", model="haiku"):
    "FILES_CHANGED: {files modified by fix Coder}
    VALIDATION_SCOPE: changed-only"
    ```
@@ -526,7 +526,7 @@ Step 3: GATE — Verify TeamDelete succeeded
 After Evaluator passes, spawn Tester for scenario-based acceptance testing (standalone agent, not a teammate — testing is sequential, not debate):
 
 ```
-Task(subagent_type="Tester"):
+Agent(subagent_type="Tester"):
 "ORIGINAL_REQUEST: {task description or issue content}
 EXECUTION_PLAN: {synthesized plan from Phase 6}
 FILES_CHANGED: {list of files from Coder output}
@@ -542,7 +542,7 @@ Design and execute scenario-based acceptance tests. Report PASS or FAIL with evi
 3. If `qa_retry_count <= 2`:
    - Spawn Coder to fix QA failures:
    ```
-   Task(subagent_type="Coder"):
+   Agent(subagent_type="Coder"):
    "TASK_ID: {task-id}
    TASK_DESCRIPTION: Fix QA test failures
    OPERATION: qa-fix
@@ -552,7 +552,7 @@ Design and execute scenario-based acceptance tests. Report PASS or FAIL with evi
    ```
    - Spawn Validator to verify fix didn't break tests:
    ```
-   Task(subagent_type="Validator", model="haiku"):
+   Agent(subagent_type="Validator", model="haiku"):
    "FILES_CHANGED: {files modified by fix Coder}
    VALIDATION_SCOPE: changed-only"
    ```
@@ -652,7 +652,7 @@ If the Coder's report includes Key Decisions with architectural significance:
 6. **Clean handoffs** - Each phase passes structured data to next; sequential Coders pass implementation summaries
 7. **Honest reporting** - Display agent outputs directly
 8. **Simplification pass** - Code refined for clarity before PR
-9. **Strict delegation** - Never perform agent work in main session. "Spawn X" means call Task tool with X, not do X's work yourself
+9. **Strict delegation** - Never perform agent work in main session. "Spawn X" means call Agent tool with X, not do X's work yourself
 10. **Validator owns validation** - Never run `npm test`, `npm run build`, or similar in main session; always delegate to Validator agent
 11. **Coder owns fixes** - Never implement fixes in main session; spawn Coder for validation failures and alignment fixes
 12. **Loop limits** - Max 2 validation retries, max 2 alignment fix iterations before escalating to user