pwn 0.5.680 → 0.5.683

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ca4257e59d3ee50dde626ddb4aabf356b578ebd2bb82a23f3b724cae214d6d09
-  data.tar.gz: 62c7c4a1b139769fe86e20ea0243f0cc2d7155eb15421dcd2e29c640b6bd838d
+  metadata.gz: 138deb9b9916fd71329a5898d6cb49ab53751e597f408177aee7682704f33500
+  data.tar.gz: 77f88100898efbff17bea33bbd2efa186afecf2776c02d3e577a71b26d1b2d3b
 SHA512:
-  metadata.gz: adca8d361a9535a64bac1804a5cf5ae04b1736bfa4ecde55be7b956dfa33d02e7f8e75827bd76164af9c103b6e1d188fa53ed6627fe38677ff1263512c202795
-  data.tar.gz: 0de0c0e020d650c893322714b7ea90a1a91c9d0a4e38c475fad8e37584711c0f8d985571e8514fcaf05fd0d8c2a02010974b44fd3fe923ca1ee89014002a8be0
+  metadata.gz: 265a38e4abaead447360e549c98f88dc0a3d3220358d0e30c9e9419d6c1b53dd558dac87b2ce093e968b53bbab9051db7032b405d5835194c33934acd64ecb4d
+  data.tar.gz: 7c0e34b2be0c96ddef980ee81493f5b2d394680ca956f0b6208cc32758fda13f212670febca563092cd5b0331ac46464bdecd0f6c88d2f7a15744f4b1cd79d48

data/.gitignore

@@ -1,5 +1,6 @@
 ._*
 .each*
+.env
 *.swp
 *.gem
 *.rbc

data/documentation/AI-Integration.md

@@ -48,7 +48,7 @@ The harness adapts to the *class* of engine, not the model name:
 | **MEMORY ranking** | relevance-ranked when a local Ollama `embed_model` is reachable, else newest-first | relevance-ranked via `PWN::MemoryIndex` (`~/.pwn/memory.idx`) |
 | **Tool schemas shipped** | all toolsets | `CORE_TOOLS` + top-K keyword matches when `ai.agent.tool_router` is on (default on). Ties also honor `tool_preference`. |
 | **Pre-pass** | none | `plan_first` numbered tool plan before first dispatch |
-| **Intent route** | always | `request_intent` + LLM/heuristic `request_kind` (statement \| question \| autonomous_goal). Short-circuits how-to/questions (text only), greetings/statements (fixed ack), pure recall, and unauthorized recon on all engines; host-evidence Qs (hostname/cwd/whoami) and only true autonomous goals get multi-step TaskSummarizer plans. Critical for ollama/openwebui |
+| **Intent route** | always | `request_intent` only (greeting / howto / recall cheap paths). There is no request type. Every other turn gets a TaskSummarizer compass and CORE_TOOLS. |
 | **Few-shot** | none | `Learning.exemplars_for(request)` splices a prior successful trace |
 | **Dispatch parsing** | strict | tolerant - Levenshtein tool-name repair + JSON5-ish arg cleanup, each repair fingerprinted into `Mistakes` |
 | **Post-answer** | `auto_introspect` (deferred by default) | `auto_introspect` **+** `fact_check_local_final` (auto `extro_verify` on CVE/version-shaped claims). Deferred after the reply; specs/cron stay inline. |

data/documentation/Agent-Tool-Registry.md

@@ -13,8 +13,8 @@ toolsets; the JSON-Schema for each tool is what the model actually sees.
 | `terminal` | `shell` | `Open3.capture3` on the host, after `PWN::AI::Agent::ToolGuard` |
 | `pwn` | `pwn_eval` | `TOPLEVEL_BINDING.eval` in the live REPL process, after `ToolGuard` |
 | `memory` | `memory_remember` · `memory_recall` · `memory_forget` · `memory_clear` · **`memory_lean`** | `PWN::Memory` → `~/.pwn/memory.json` |
-| `skills` | `skill_list` · `skill_view` · `skill_create` · `skill_add_reference` · `skill_delete` · `skill_migrate_legacy` | `~/.pwn/skills/<name>/SKILL.md` (**[agentskills.io](https://agentskills.io) spec**; legacy flat `*.md` auto-migrated) |
-| `sessions` | `sessions_list` · `sessions_view` · `sessions_current` · `sessions_delete` · `sessions_stats` · **`sessions_lean`** | `PWN::Sessions` → `~/.pwn/sessions/` |
+| `skills` | **`skills_recall`** · `skill_list` · `skill_view` · `skill_create` · `skill_add_reference` · `skill_delete` · `skill_migrate_legacy` | `~/.pwn/skills/<name>/SKILL.md` (**[agentskills.io](https://agentskills.io) spec**; legacy flat `*.md` auto-migrated) |
+| `sessions` | **`session_recall`** · `sessions_list` · `sessions_view` · `sessions_current` · `sessions_delete` · `sessions_stats` · **`sessions_lean`** | `PWN::Sessions` → `~/.pwn/sessions/` |
 | `learning` | `learning_note_outcome` · `learning_reflect` · `learning_distill_skill` · `learning_stats` · `learning_outcomes` · `learning_consolidate` · `learning_reset` · `learning_auto_introspect_toggle` · **`learning_gc_stores`** · **`learning_purge_noise`** · **`mistakes_list`** · **`mistakes_record`** · **`mistakes_resolve`** · **`mistakes_reset`** · **`mistakes_lean`** · **`reward_judge`** · **`reward_prm`** · **`reward_sentinel`** · **`reward_preferences`** · **`reward_export_dpo`** · **`reward_warm_sentinel`** · **`reward_scrub_preferences`** · **`reward_preference_balance`** · **`curriculum_practice`** · **`curriculum_train`** · **`curriculum_hindsight`** · **`curriculum_offline_judge`** · **`curriculum_preference_balance`** | `PWN::AI::Agent::Learning` + `Mistakes` + `Reward` + `Curriculum` → `~/.pwn/learning.jsonl` + `~/.pwn/mistakes.json` + `~/.pwn/preferences.jsonl` + `~/.pwn/curriculum/` + `~/.pwn/finetune/` |
 | `reward` | **`reward_generator_mix`** | `PWN::AI::Agent::Reward.generator_mix` → online preference source-mix controller (`preferences.jsonl`) |
 | `curriculum` | **`curriculum_practice_kpi`** | `PWN::AI::Agent::Curriculum.practice_kpi` → `~/.pwn/curriculum_kpi.jsonl` |
@@ -56,8 +56,9 @@ user request through as `relevance:`, `Registry.definitions` shrinks the
 pool to:
 
 ```text
-CORE_TOOLS  = shell · pwn_eval · memory_remember · memory_recall
-              mistakes_record · mistakes_resolve · learning_note_outcome
+CORE_TOOLS / DEFAULT_PREFERENCE (same list):
+memory_recall · session_recall · skills_recall · pwn_eval · shell
+mistakes_record · mistakes_resolve · learning_note_outcome · memory_remember
             + top-K keyword-ranked matches for THIS request
               (ties break on Metrics per-engine success_rate, then
                ai.agent.tool_preference)
@@ -80,7 +81,7 @@ When keyword fit and other rank scores tie, the registry prefers this
 default order:
 
 ```text
-memory_recall · sessions_view · pwn_eval · shell
+memory_recall · session_recall · skills_recall · pwn_eval · shell
 mistakes_record · mistakes_resolve · learning_note_outcome · memory_remember
 ```
 
@@ -88,9 +89,12 @@ Set `ai.agent.tool_preference` in `~/.pwn/pwn.yaml`, or pass `order:` /
 `preference:` into `Registry.definitions`, `.rank`, or `.apply_preference`.
 An explicit empty list turns preference off (no Env / default fallback).
 
+Learned facts and this session are injected (MEMORY / RECENT TURNS), not
+first tools. Preference then lists `pwn_eval` before `shell`. `sessions_view`
+is not a CORE tool. There is no separate ACT_PREFERENCE.
+
 Keyword fit stays the primary signal. Preference is a smaller bonus plus a
-stable sort after the router slims the pool, so `memory_recall` wins a
-tie against `shell` without hiding a better keyword match.
+stable sort after the router slims the pool.
 
 `Policy` uses the same list when it suggests a next action in the prompt.
 

data/documentation/Configuration.md

@@ -123,8 +123,7 @@ ai:
     task_summary_every: 5          # When task_summary_verbose: emit Progress every N completed tools.
     task_summary_interval_s: 8.0   # When verbose: also emit when this many seconds elapsed.
     task_summary_verbose: false    # Mid-flight Progress/Finished lines (default: only plan + about_to).
-    task_summary_llm: true         # LLM tangible-task decompose for autonomous goals (default on). false = offline fallback.
-    request_kind_llm: ~            # LLM request_kind classifier (statement|question|autonomous_goal). nil = follow task_summary_llm.
+    task_summary_llm: true         # LLM tangible-task decompose (default on). false = offline fallback.
     max_depth: 3                   # Recursion guard: how many levels deep agent_ask/agent_debate sub-agents may spawn sub-agents.
     auto_introspect: true          # Run Learning.auto_introspect (outcome logging + lesson mining) after every final answer.
     auto_extrospect: true          # Ambient baseline after every final answer (host/repo/env ONLY - never launches burpsuite/zaproxy/msf/gqrx). Sense tools stay on-demand.
@@ -134,9 +133,10 @@ ai:
     shell_bash: false              # true -> run shell via bash -lc. Default is /bin/sh.
     plan_first: ~                  # Plan-then-act pre-pass. nil = auto (true when ai.active is ollama or openwebui).
     tool_router: ~                 # Dynamic tool-set slimming. nil = auto (true for ollama / openwebui).
-    tool_preference:               # Operator-tunable tool order. Rank bonus + Policy suggested-action list.
+    tool_preference:               # Same order as CORE_TOOLS. Current session is injected; then memory_recall, session_recall, skills_recall, pwn_eval, shell.
       - memory_recall
-      - sessions_view
+      - session_recall
+      - skills_recall
       - pwn_eval
       - shell
       - mistakes_record
@@ -308,8 +308,7 @@ PWN::Config.refresh_env
 | `ai.agent.task_summary_every` | Integer | `5` | `TaskSummarizer.every_n` | Verbose progress cadence (tools). |
 | `ai.agent.task_summary_interval_s` | Float | `8.0` | `TaskSummarizer.interval_s` | Verbose progress cadence (seconds). |
 | `ai.agent.task_summary_verbose` | Boolean | `false` | `TaskSummarizer.verbose?` | Emit mid-flight `Progress:` / `Finished:` lines; default keeps only plan + about_to. |
-| `ai.agent.task_summary_llm` | Boolean \| `nil` | `nil` (on) | `TaskSummarizer.llm_plan_enabled?` | LLM tangible-task decomposition for autonomous goals. `false` forces offline generic fallback (tests / air-gap). |
-| `ai.agent.request_kind_llm` | Boolean \| `nil` | `nil` (follow `task_summary_llm`) | `TaskSummarizer.llm_kind_enabled?` / `request_kind` | LLM classifier for `statement` \| `question` \| `autonomous_goal`. Cheap intents and host-evidence heuristics still win first; `false` is heuristic-only. |
+| `ai.agent.task_summary_llm` | Boolean \| `nil` | `nil` (on) | `TaskSummarizer.llm_plan_enabled?` | LLM tangible-task decomposition. `false` forces offline generic fallback (tests / air-gap). |
 | `ai.agent.max_depth` | Integer | `3` | `PWN::AI::Agent::Swarm` | Recursion guard for `agent_ask` / `agent_debate` sub-agents spawning sub-agents. |
 | `ai.agent.auto_introspect` | Boolean | `true` | `PWN::AI::Agent::Learning.auto_introspect` | Run outcome logging + lesson mining after every final answer. Toggle live via `learning_auto_introspect_toggle`. |
 | `ai.agent.auto_extrospect` | Boolean | `true` | `PWN::AI::Agent::Extrospection.auto_extrospect` | Ambient baseline after every final answer (`AUTO_SECTIONS` = host/repo/env only; never spawns GUI/JVM tools). Sense tools (`intel`/`verify`/`watch`/`rf_tune`/`observe`) stay on-demand. Toggle live via `extro_auto_toggle`. |
@@ -318,7 +317,7 @@ PWN::Config.refresh_env
 | `ai.agent.shell_bash` | Boolean | `false` | `PWN::AI::Agent::ToolGuard.shell_bash?` | When true, `shell` runs via `bash -lc` so bash-only syntax is allowed. Default is POSIX `/bin/sh` and bashisms are rejected with a rewrite hint. |
 | `ai.agent.plan_first` | Boolean \| `nil` | `nil` (auto: `true` when `ai.active` is `ollama` or `openwebui`) | `PWN::AI::Agent::Loop.plan_first` | Plan-then-act pre-pass: the model must emit a numbered tool plan (as an assistant message) *before* it may dispatch anything. Cheap chain-of-thought scaffolding for local models. |
 | `ai.agent.tool_router` | Boolean \| `nil` | `nil` (auto: `true` for `ollama` / `openwebui`) | `PWN::AI::Agent::Registry.definitions` | Dynamic tool-set slimming: expose only `Registry::CORE_TOOLS` + the top-K keyword-relevant schemas for *this* request. Ties break on historical `Metrics` success rate, then `ai.agent.tool_preference`. |
-| `ai.agent.tool_preference` | Array\<String\> | `memory_recall`, `sessions_view`, `pwn_eval`, `shell`, `mistakes_record`, `mistakes_resolve`, `learning_note_outcome`, `memory_remember` | `PWN::AI::Agent::Registry.preference_order` / `.rank` / `.apply_preference`, `Policy` | Operator-tunable tool order. Keyword fit stays primary; this list is a rank bonus and the default action list Policy quotes in the prompt. Explicit empty list disables preference. |
+| `ai.agent.tool_preference` | Array\<String\> | `memory_recall`, `session_recall`, `skills_recall`, `pwn_eval`, `shell`, `mistakes_record`, `mistakes_resolve`, `learning_note_outcome`, `memory_remember` | `PWN::AI::Agent::Registry.preference_order` / `.rank` / `.apply_preference`, `Policy` | Same order as `CORE_TOOLS`. Current session is injected; then `memory_recall`, `session_recall`, `skills_recall`, `pwn_eval`, `shell`. Explicit empty list disables preference. |
 | `ai.agent.defer_introspect` | Boolean | `true` | `PWN::AI::Agent::TurnFinalizer` | Run `Learning.auto_introspect` on a background thread after the user-visible reply. Specs and cron stay inline. |
 | `ai.agent.prompt_cache` | Boolean | `true` | `PWN::AI::Agent::PromptCache` | Engine-native prefix cache. Anthropic uses `cache_control`; OpenAI uses `prompt_cache_key`; Grok uses `x-grok-conv-id`; Gemini splits `systemInstruction`. Ollama and Open WebUI have no native prefix-cache field. |
 | `ai.agent.local_introspect` | Symbol | `failure_only` | `PWN::AI::Agent::Learning.auto_introspect` | End-of-turn introspect policy for local engines: `always` · `failure_only` · `every_n` (with `introspect_every_n`). |

data/documentation/How-PWN-Works.md

@@ -28,8 +28,8 @@ hardware).
 | Module | Role |
 |---|---|
 | `Loop` | plan → **TaskSummarizer** briefs → dispatch tool_calls → observe → repeat until final answer; tightens runway when recent turns exhausted the budget |
-| **`TaskSummarizer`** | Executive UX: LLM/heuristic `request_kind` gates plans (only autonomous goals, including host-evidence Qs) - English tasks primary - `emit_plan!` · `about_to` as `task k/n` + via tools · Loop `plan_context` injection · clearer `plan_idx` advance |
-| `Registry` | JSON-Schema function definitions grouped into 13 **toolsets** · **85 tools** · `tool_preference` order (default memory_recall-first) |
+| **`TaskSummarizer`** | Executive UX: every request gets an English task compass (`emit_plan!` · `about_to` as `task k/n`) — no statement/question/goal type |
+| `Registry` | JSON-Schema function definitions grouped into 13 **toolsets** · **87 tools** · `CORE_TOOLS` = `DEFAULT_PREFERENCE` (`memory_recall` · `session_recall` · `skills_recall` · `pwn_eval` · `shell`) |
 | `Dispatch` / `Result` | execute a tool, capture stdout/value/error/duration |
 | `PromptBuilder` | inject MEMORY / SKILLS / LEARNING / **KNOWN MISTAKES + FIXES** / METRICS / **POLICY** / EXTROSPECTION / RECENT TURNS |
 | `Metrics` · `Learning` · `Reflect` · **`Policy`** | **introspection** - how well am I doing? (Policy is live Q / REINFORCE, advisory rank only) |

data/documentation/Reinforcement-Learning.md

@@ -169,7 +169,7 @@ This table is the live control list. Track the outcomes, not source comments.
     :max_iters: 75           # hard cap; budget pressure may lower effective value
     :defer_introspect: true  # post-answer Learning after the user-visible reply
     :prompt_cache: true      # engine-native prefix cache (not ollama / openwebui)
-    :tool_preference: [memory_recall, sessions_view, pwn_eval, shell, mistakes_record, mistakes_resolve, learning_note_outcome, memory_remember]
+    :tool_preference: [memory_recall, session_recall, skills_recall, pwn_eval, shell, mistakes_record, mistakes_resolve, learning_note_outcome, memory_remember]
 ```
 
 ## Cron self-improvement

data/documentation/diagrams/agent-tool-registry.svg

@@ -29,7 +29,7 @@
 <text xml:space="preserve" text-anchor="middle" x="105.12" y="-521.82" font-family="sans-Serif" font-size="12.00" fill="#0f172a">JSON&#45;Schema function defs</text>
 <text xml:space="preserve" text-anchor="middle" x="105.12" y="-507.57" font-family="sans-Serif" font-size="12.00" fill="#0f172a">rank(query) · CORE_TOOLS</text>
 <text xml:space="preserve" text-anchor="middle" x="105.12" y="-493.32" font-family="sans-Serif" font-size="12.00" fill="#0f172a">tool_router → CORE + top&#45;K</text>
-<text xml:space="preserve" text-anchor="middle" x="105.12" y="-479.07" font-family="sans-Serif" font-size="12.00" fill="#0f172a">tool_preference (memory_recall first)</text>
+<text xml:space="preserve" text-anchor="middle" x="105.12" y="-479.07" font-family="sans-Serif" font-size="12.00" fill="#0f172a">tool_preference (pwn_eval then shell)</text>
 </g>
 <!-- terminal -->
 <g id="node2" class="node">

data/documentation/diagrams/dot/agent-tool-registry.dot

@@ -9,7 +9,7 @@ digraph "PWN_Agent_Tool_Registry" {
         shape=box, penwidth=1.3, color="#334155", fontcolor="#0f172a"];
   edge [color="#94a3b8", penwidth=1.1, arrowsize=0.7];
 
-  Registry [label="Registry\nJSON-Schema function defs\nrank(query) · CORE_TOOLS\ntool_router → CORE + top-K\ntool_preference (memory_recall first)", fillcolor="#c4b5fd", fontsize=12, penwidth=2];
+  Registry [label="Registry\nJSON-Schema function defs\nrank(query) · CORE_TOOLS\ntool_router → CORE + top-K\ntool_preference (pwn_eval then shell)", fillcolor="#c4b5fd", fontsize=12, penwidth=2];
 
   subgraph cluster_ts {
     label="Toolsets"; fontcolor="#a7f3d0"; style=rounded;

data/documentation/diagrams/dot/task-summarizer.dot

@@ -1,6 +1,6 @@
 digraph "PWN_TaskSummarizer" {
   graph [
-    label=<<B>PWN::AI::Agent::TaskSummarizer - Request kind + executive briefs</B><BR/><FONT POINT-SIZE="11" COLOR="#94a3b8">request_kind → (statement|question: no plan) · (autonomous_goal: emit_plan! → about_to → tools)</FONT>>,
+    label=<<B>PWN::AI::Agent::TaskSummarizer - executive briefs</B><BR/><FONT POINT-SIZE="11" COLOR="#94a3b8">every request → emit_plan! → about_to → tools (no request type)</FONT>>,
     labelloc=t, fontsize=20, fontname="Helvetica",
     rankdir=TB, splines=spline, nodesep=0.55, ranksep=0.9,
     bgcolor="#0f172a", fontcolor="#e2e8f0", pad=0.6, newrank=true, compound=true
@@ -11,8 +11,8 @@ digraph "PWN_TaskSummarizer" {
         fontcolor="#cbd5e1", penwidth=1.3, arrowsize=0.8];
 
   User  [label="User request", fillcolor="#7dd3fc"];
-  Loop  [label="Loop.run\nts_state = TaskSummarizer.fresh\nrequest_kind + request_intent", fillcolor="#c4b5fd"];
-  Kind  [label="request_kind (LLM + heuristic)\nstatement | question | autonomous_goal\nhost-evidence Qs → goal", fillcolor="#fcd34d", penwidth=2];
+  Loop  [label="Loop.run\nts_state = TaskSummarizer.fresh\nrequest_intent only", fillcolor="#c4b5fd"];
+  Kind  [label="no request type\nevery turn is a goal", fillcolor="#fcd34d", penwidth=2];
   {rank=same; User; Loop; Kind}
 
   subgraph cluster_nogoal {

data/documentation/pwn-ai-Agent.md

@@ -126,19 +126,11 @@ full `Loop.run` under a persona overlay) that share a JSONL bus. See
 ## Task summaries (long autonomous turns)
 
 `PWN::AI::Agent::TaskSummarizer` keeps the TUI readable during multi-step work.
-First it classifies the request with `request_kind` (LLM when enabled, else
-heuristics / `request_intent`):
-
-| Kind | Example | Task breakdown |
-|---|---|---|
-| `statement` | "FYI the build is green." | None - brief ack only |
-| `question` | "what is the default GQRX port?" / "how to ...?" | None - concise answer only |
-| `autonomous_goal` | "refactor Loop.run and run rubocop" / "what is my hostname?" | **Required** ordered work units (each may use one or more tools) |
+There is no request type. Every request gets an English task compass.
 
 | Surface | When | Content |
 |---|---|---|
-| `request_kind` | User submit | `statement` \| `question` \| `autonomous_goal` |
-| `emit_plan!` | Autonomous goals only | **Full** goal + ordered plain-English tangible tasks (each may need many tools) |
+| `emit_plan!` | User submit | **Full** goal + ordered plain-English tangible tasks (each may need many tools) |
 | `about_to` | Before each tool batch | **Primary:** `task k/n: <english>` - **secondary:** `via shell×2 (search)` (not raw argv) |
 | `plan_context` / `active_task_prompt` | Into Loop messages | Same English tasks steer tool choice (not TUI-only) |
 | `record!` | After each tool | Advances `plan_idx`; emits English advancement brief when the index moves; verbose progress only if `task_summary_verbose` |
@@ -163,8 +155,7 @@ task_summary: true              # master switch (default on)
 task_summary_every: 5           # verbose progress every N tools
 task_summary_interval_s: 8.0    # or every N seconds (verbose)
 task_summary_verbose: false     # mid-flight Progress: lines
-task_summary_llm: true          # LLM task decompose for autonomous goals (default on)
-request_kind_llm: null          # null = follow task_summary_llm; false = heuristic-only
+task_summary_llm: true          # LLM task decompose (default on)
 max_iters: 75                   # budget pressure may lower the effective cap (stricter on local engines)
 ```
 
@@ -206,7 +197,7 @@ max_iters: 75                   # budget pressure may lower the effective cap (s
 | `reward_llm_timeout` | `12` | seconds for the cheap ORM chat (clamped 2..30) |
 | `verify_as_reward` | `nil` (auto) | browser-grounded claim sample policy |
 | `local_introspect` | `:failure_only` | ollama / openwebui end-of-turn introspect policy |
-| `tool_preference` | memory_recall-first list | Rank bonus + Policy suggested-action order |
+| `tool_preference` | same list as CORE_TOOLS (`memory_recall`, `session_recall`, `skills_recall`, `pwn_eval`, `shell`) | Rank bonus + Policy suggested-action order |
 | `defer_introspect` | `true` | Post-answer Learning on a background thread |
 | `prompt_cache` | `true` | Engine-native prefix cache (not Ollama / Open WebUI) |
 
@@ -219,28 +210,18 @@ Full detail: [Reinforcement Learning](Reinforcement-Learning.md).
 
 [← Home](Home.md)
 
-## Intent routing (kind + fine-grained intent)
-
-Every user turn is classified on two layers:
-
-1. **`request_kind`** - `statement` | `question` | `autonomous_goal` (TaskSummarizer is the
-   single source of truth; `Loop.request_kind` delegates). Classification order:
-   injected label (tests) → cheap `request_intent` short-circuits (greeting/howto/recall/recon) →
-   strong agent-do / host-evidence heuristics → **LLM classify** via `chat_for_kind`
-   (when `ai.agent.request_kind_llm` is on / follows `task_summary_llm`) → offline
-   `heuristic_request_kind`. Only autonomous goals receive a multi-step tangible-task plan.
-2. **`request_intent`** - fine-grained route for cheap short-circuits and recon guard.
-
-| Intent | Kind | Example | Behavior |
-|--------|------|---------|----------|
-| How-to | question | "how to do a ping sweep of a subnet using hping3?" | Short explanation with example commands only. No tools and no multi-step plan. |
-| Question | question | "what is the default GQRX remote-control port?" | Concise answer. No multi-step task breakdown. |
-| Host evidence | autonomous_goal | "what is my hostname?" / "excellent - what is my hostname?" | Needs a live local lookup (hostname, cwd, whoami, IP, ...). Treated as a goal so tools run; not text-only Q&A. |
-| Greeting | statement | "Howdy, it's cloudy." / "hi" | Fixed short ack that the system is ready. No tools, no LLM, and no weather echo such as "noted, cloudy out there." |
-| Statement | statement | "FYI the build is green." | Brief note. No multi-step task plan. |
-| Recall | question | "what did I just say?" / "how did you respond?" | Cheap prior-turn answer from the session transcript. No plan_first and no multi-tool archaeology. |
-| Live recon | autonomous_goal | "using hping3 what live hosts can you find in this subnet?" | Needs clear in-scope / authorized engagement wording, or set `ai.agent.recon_authorized=true`. Otherwise the agent refuses and points you at the how-to form. |
-| Act | autonomous_goal | "refactor Loop.run and run rubocop" | Normal multi-step agent work: decompose into ordered work units, each may use one or more tools. |
+## Intent routing
+
+There is no request type. Greeting / howto / recall still use `request_intent`
+for cheap short-circuits. Everything else is a goal: TaskSummarizer compass + CORE_TOOLS.
+
+| Intent | Example | Behavior |
+|--------|---------|----------|
+| How-to | "how to do a ping sweep of a subnet using hping3?" | Short explanation with example commands only. No tools. |
+| Greeting | "Howdy, it's cloudy." / "hi" | Fixed short ack. No tools, no LLM, no weather echo. |
+| Recall | "what did I just say?" / "how did you respond?" | Cheap prior-turn answer from the session transcript. |
+| Live recon | "using hping3 what live hosts can you find in this subnet?" | Needs in-scope / authorized wording, or `ai.agent.recon_authorized=true`. |
+| Goal | "refactor Loop.run and run rubocop" / "what color is a cherry" | Task compass + CORE_TOOLS. There is no statement/question type. |
 
 The `shell` tool also blocks hping3 / nmap-style sweep commands when recon is
 not authorized. On how-to asks, memory SOPs about repo rubocop/rake hygiene are