summaryrefslogtreecommitdiff
path: root/internal/role/role.go
diff options
context:
space:
mode:
authorClaude Sonnet 5 <noreply@anthropic.com>2026-07-03 23:01:50 +0000
committerClaude Sonnet 5 <noreply@anthropic.com>2026-07-03 23:01:50 +0000
commit787b7fb1aed92c2b701724a7741576053b93cccb (patch)
tree372aac85a10093a20ce3152e2c11fcfe20bb3e9d /internal/role/role.go
parent1f203a7ac0efad15ec3fc0a4c5b335ad7073a52f (diff)
feat(role): add versioned role configs + escalation ladder + scheduler (Phase 5)
Two parts: Part A (fixes a gap from Phase 1): Groq/OpenRouter/OpenAI were documented in docs/api-keys-setup.md as usable once configured, but nothing actually constructed runners for them. internal/cli/cloudrunners.go consolidates anthropic/google/groq/openrouter/openai NativeRunner construction into one table-driven registerCloudRunners() helper, replacing the two hand-written per-provider blocks in serve.go/run.go. Groq/OpenRouter/OpenAI reuse openaicompat (no new adapter code) at SandboxKind: "docker". Part B: the token-husbanding harness's core routing mechanism. - internal/role: RoleConfig/Tier/Rung -- a role's system prompt and a multi-tier (provider, model) escalation ladder, versioned via config_json. - storage: new role_configs table (draft/active/retired, UNIQUE(role, version)) with transactional activate-retires-prior-active semantics; new executions.escalation_rung column. - task.AgentConfig.Role string -- purely additive; every existing task shape (Agent.Role == "") is unaffected, proven by TestPool_Execute_NonRoleTask_Unaffected plus the full pre-existing suite passing unchanged. - executor.Pool.execute(): role-typed tasks with no Agent.Type yet resolve tier 0 of their active ladder (round-robin across multi-candidate tiers, skipping rate-limited providers, falling back to soonest-clearing) before the existing pickAgent/Classifier path runs; SystemPrompt applies to Agent.SystemPromptAppend. Already-resolved role tasks (scheduler resubmits) get their escalation_rung re-derived read-only via findTierIndex. - internal/scheduler: polls role-typed FAILED tasks, retries at the same rung under MaxRetries or escalates to the next tier's first candidate when budget.Accountant.Allow() permits (emitting event.KindEscalated), else leaves the task FAILED with a final:true KindEscalated event. An in-memory per-execution-ID "handled" set keeps the poll loop convergent. Started by `serve` only, config knob [scheduler].poll_interval_seconds. - internal/api: POST/GET /api/roles/{role}/versions, POST /api/roles/{role}/activate -- unauthenticated, matching the existing projects/tasks REST endpoints' auth posture (only chatbot MCP, agent MCP, and WebSocket are api_token-gated in this codebase today). Documented as stored-but-not-yet-enforced (CLAUDE.md Design Debt, matching how task.Priority/RetryConfig are already documented): RoleConfig.Tools/ SandboxKind don't affect dispatch yet; DefaultBudgetUSD is read narrowly as the scheduler's escalation cost estimate, not enforced at initial dispatch; scheduler escalation always targets Candidates[0] (no round-robin, unlike initial-dispatch tier-0 resolution); the scheduler's dedupe is per-process and resets on restart (idempotent, harmless). go build/vet/test -race -count=1 all pass, 21 packages. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V1moSNCJRcP6kykA4tyUSs
Diffstat (limited to 'internal/role/role.go')
-rw-r--r--internal/role/role.go73
1 files changed, 73 insertions, 0 deletions
diff --git a/internal/role/role.go b/internal/role/role.go
new file mode 100644
index 0000000..f65edba
--- /dev/null
+++ b/internal/role/role.go
@@ -0,0 +1,73 @@
+// Package role defines the per-role configuration shape (system prompt, tool
+// allowlist, sandbox kind, budget, and — the part actually driven this phase —
+// a multi-tier provider/model escalation ladder) used by the token-husbanding
+// harness. A RoleConfig is stored as a versioned row in storage's
+// role_configs table (see internal/storage/roleconfig.go); internal/executor
+// resolves tier 0 of a role's EscalationLadder for the initial dispatch of a
+// role-typed task, and internal/scheduler walks the ladder on failure
+// (retry-then-escalate).
+//
+// Not everything on RoleConfig is enforced yet. See the field doc comments
+// below and CLAUDE.md's Design Debt section for what's stored-but-not-wired,
+// matching how this codebase already documents task.Priority/RetryConfig.
+package role
+
+// RoleConfig is one version of a role's configuration.
+type RoleConfig struct {
+ // Role names the role this config applies to (e.g. "coder", "reviewer").
+ Role string `json:"role"`
+
+ // SystemPrompt is appended to Agent.SystemPromptAppend when a task
+ // dispatches through this role (internal/executor.Pool.execute()). Wired.
+ SystemPrompt string `json:"system_prompt,omitempty"`
+
+ // Tools, SandboxKind, and DefaultBudgetUSD are stored and round-tripped
+ // through the API/storage layer but are NOT YET enforced by the executor:
+ // guardrails/tool availability are still applied uniformly by
+ // NativeRunner (Phase 3) regardless of role, and SandboxKind here has no
+ // effect on which sandbox.Sandbox implementation a runner picks (that's
+ // still purely provider-driven, see NativeRunner.SandboxKind). A later
+ // phase should thread these through. DefaultBudgetUSD is read by
+ // internal/scheduler as the estimated cost passed to
+ // budget.Accountant.Allow() when considering an escalation, which is a
+ // narrow, specific use — not general per-role budget enforcement at
+ // dispatch time.
+ Tools []string `json:"tools,omitempty"`
+ SandboxKind string `json:"sandbox_kind,omitempty"`
+ DefaultBudgetUSD float64 `json:"default_budget_usd,omitempty"`
+
+ // EscalationLadder is the ordered list of tiers a role-typed task climbs
+ // on repeated failure. Tier 0 is used for the initial dispatch.
+ EscalationLadder []Tier `json:"escalation_ladder,omitempty"`
+}
+
+// Tier is one rung-group in an escalation ladder: a set of candidate
+// (provider, model) rungs to choose among, plus how many attempts may be made
+// at this tier before the scheduler escalates to the next one.
+type Tier struct {
+ Candidates []Rung `json:"candidates"`
+ // SelectionMode is "round_robin" (default, when empty) or "single" (always
+ // use Candidates[0]).
+ SelectionMode string `json:"selection_mode,omitempty"`
+ // MaxRetries is the number of attempts allowed at this tier before the
+ // scheduler escalates to the next tier. 0 means "escalate immediately
+ // after the first failure at this tier" (no retries held here).
+ MaxRetries int `json:"max_retries"`
+}
+
+// Rung is a single (provider, model) pair. Provider must match a registered
+// executor runner key: "local", "anthropic", "google", "groq", "openrouter",
+// "openai".
+type Rung struct {
+ Provider string `json:"provider"`
+ Model string `json:"model"`
+}
+
+// EffectiveSelectionMode returns t.SelectionMode, defaulting to
+// "round_robin" when unset.
+func (t Tier) EffectiveSelectionMode() string {
+ if t.SelectionMode == "" {
+ return "round_robin"
+ }
+ return t.SelectionMode
+}