context + repl + config: per-class system_prompt override (closes #86)

Small local models follow precise structured instructions better than
natural language. Per-routing-class system_prompt override gives them
tighter instructions for THAT request while preserving ambient context.

Changes:

- Context:to_messages(opts) — opts.system_prompt_override REPLACES
  the base system_prompt for THIS render only (state unchanged).
  Dynamic blocks ([background], [project], [earlier summary], NORRIS
  suffix) still compose on top. opts is optional; nil-safe for old
  callers.

- repl.lua ask_ai — captures req_class from router.classify_model
  (already returned by Phase 5; previously discarded after the
  status line). Looks up config.routing.system_prompts[req_class];
  passes as opts.system_prompt_override to ctx:to_messages each
  iteration of the tool-sub-loop.

- Gating: override fires only when routing.auto is on (no class ->
  no override). If system_prompts[class] absent for a class, fall
  through to the default system_prompt (no surprise).

- Norris unaffected: safety.norris_step builds its own messages
  array; doesn't go through this path.

- config.lua gains a commented-out example showing routing.system_
  prompts with the code/default examples from the FR body.

Smoke verified:
  - 12-case context.lua unit test: opts nil/absent/present, override
    replaces base, dynamic blocks still compose, state unchanged
    after call, Norris-mode coexistence (suffix still present;
    background still suppressed).
  - E2E against cloud broker with routing.system_prompts.code set:
    triple-backtick prompt -> code class -> override fires; model
    emits terse code-only output. Non-code prompt -> default class
    -> no override -> normal verbose-ish reply.

Regression: test_safety 87/87, test_router_model 31/31, repl loads.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

config.lua

@@ -240,6 +240,26 @@ return {
     --     -- Toggle at runtime with :fallback on / :fallback off.
     --     fallback       = false,           -- default off (cost-safety)
     --     fallback_model = "cloud",
+    --
+    --     -- Issue #86: per-class system_prompt override. When the
+    --     -- classified request falls into a class with an entry here,
+    --     -- the BASE system_prompt is REPLACED for that one request
+    --     -- (dynamic blocks — [background], [project], [earlier
+    --     -- summary], NORRIS suffix — still compose on top). Mostly
+    --     -- useful for tightening small local models' instruction
+    --     -- adherence. Default {} (no override).
+    --     system_prompts = {
+    --         code = [[You are a code assistant. Rules:
+    -- 1. Output ONLY the requested code or command.
+    -- 2. No prose explanation unless explicitly asked.
+    -- 3. Wrap shell commands in CMD: prefix.
+    -- 4. Max response: 200 tokens.]],
+    --         default = [[You are a shell assistant.
+    -- Output shell commands as: CMD: <command>
+    -- Output answers as single short sentences.
+    -- Do not ask clarifying questions.]],
+    --         -- reasoning routes to cloud; no override usually needed
+    --     },
     -- },
 
     -- ── Phase 5 context summarization on sliding-window eviction.

context.lua

@@ -228,8 +228,16 @@ The user will be prompted to confirm destructive actions; expect their
 verdict in the next turn as a synthesized "[aish] ... skipped by user"
 message if they declined.]]
 
-function Context:to_messages()
+function Context:to_messages(opts)
-    local sys_content = self.system_prompt
+    -- Phase 10 (#86): per-call system_prompt_override. Replaces the
+    -- BASE system_prompt for THIS render only (state unchanged); the
+    -- dynamic blocks ([background], [project], [earlier summary],
+    -- NORRIS suffix) still compose on top. Used by ask_ai's routing
+    -- path when cfg.routing.system_prompts[class] is set — gives
+    -- small local models tighter instructions while preserving
+    -- ambient memory/project context.
+    local sys_content = (opts and opts.system_prompt_override)
+                        or self.system_prompt
     -- Phase 4 [background] memory block + Phase 6 [project] file-tree
     -- block + Phase 5 [earlier summary] block. All suppressed during
     -- Norris (R-C1 / R-C4 — avoid redundant tokens per planning

repl.lua

@@ -978,13 +978,27 @@ function M.run(config)
         -- tool-sub-loop; active_name/active_cfg are NOT mutated so the
         -- user's :model selection survives the request.
         local req_name, req_cfg = active_name, active_cfg
+        local req_class
         if config.routing and config.routing.auto then
             local routed, class = router.classify_model(text, config)
+            req_class = class
             if routed and config.models[routed] and routed ~= active_name then
                 renderer.status(("routed to %s (%s class)"):format(routed, class))
                 req_name, req_cfg = routed, config.models[routed]
             end
         end
+        -- Phase 10 (#86): per-class system_prompt override. Tightens
+        -- instruction adherence on small local models. Lookup is
+        -- gated on auto-routing being on (no class -> no override).
+        -- When the override is set, ctx:to_messages REPLACES the base
+        -- system_prompt for THIS render only; dynamic blocks
+        -- ([background], [project], [earlier summary], NORRIS suffix)
+        -- still compose on top. Norris path is unaffected — it builds
+        -- its own messages array inside safety.norris_step.
+        local sys_override = config.routing
+                         and config.routing.system_prompts
+                         and req_class
+                         and config.routing.system_prompts[req_class]
 
         local depth = 0
         local final_resp = ""
@@ -994,7 +1008,9 @@ function M.run(config)
             local text_parts      = {}
             local tool_calls_seen = {}
             local redact_mode     = secrets_mode_for(req_cfg)
-            local scrubbed_msgs   = scrub_messages(ctx:to_messages(), redact_mode)
+            local scrubbed_msgs   = scrub_messages(
+                ctx:to_messages({ system_prompt_override = sys_override }),
+                redact_mode)
             -- Streaming rehydrator wraps the on_delta so the user sees real
             -- values; text_parts accumulates the REHYDRATED chunks so
             -- final_resp (used for CMD: / DELEGATE: extraction) is plain.