refactor(athena): strip audit bias from council member prompt + add intent addendums + write boundary hook + relax restrictions

src/agents/athena/council-intent-addendums.ts

@@ -0,0 +1,167 @@
+export const COUNCIL_INTENT_ADDENDUMS: Record<string, string> = {
+  AUDIT: `
+## Analysis Intent: AUDIT
+
+You are conducting an **audit** — your goal is to find discrete issues, risks, or violations.
+
+**Focus:**
+- Search for problems, anti-patterns, security risks, correctness issues, or violations of stated requirements
+- Each finding must be a distinct, actionable item with concrete evidence
+- Severity determines priority: critical (blocks/breaks), high (significant risk), medium (should fix), low (nice to fix)
+- For each finding, provide the specific location (reference, section, or component where it occurs)
+- State your confidence: high (clear evidence), medium (likely but needs verification), low (suspicion, investigate further)
+
+**Required output fields per finding:**
+- Title, Severity (critical/high/medium/low), Location, Confidence (high/medium/low)
+- Issue description, Supporting evidence, Suggested fix
+
+**Structure your response as:**
+\`\`\`
+<COUNCIL_MEMBER_RESPONSE>
+## Finding 1: [Title]
+- **Severity**: high
+- **Location**: [specific reference — e.g. component, section, endpoint, rule]
+- **Confidence**: high
+- **Issue**: [what is wrong and why it matters]
+- **Evidence**: [concrete reference, snippet, or observation that proves the issue]
+- **Suggested Fix**: [actionable recommendation]
+
+## Finding 2: [Title]
+...
+
+## Summary
+[Total findings by severity. Overall risk assessment with confidence levels.]
+</COUNCIL_MEMBER_RESPONSE>
+\`\`\``,
+
+  EVALUATE: `
+## Analysis Intent: EVALUATE
+
+You are conducting an **evaluation** — your goal is to compare options against criteria and surface tradeoffs.
+
+**Focus:**
+- Identify the options available (explicit or implied by the question)
+- Define evaluation criteria relevant to the context (cost, complexity, performance, maintainability, risk, etc.)
+- Assess each option against each criterion with evidence, not assumptions
+- Surface tradeoffs clearly — where one option wins, another likely loses
+- If the question implies a recommendation is needed, provide one with conditions
+
+**Required output fields:**
+- Options identified, Criteria used, Per-option assessment, Tradeoff summary
+- Conditional recommendation (if decision is required), Confidence per assessment
+
+**Structure your response as:**
+\`\`\`
+<COUNCIL_MEMBER_RESPONSE>
+## Options Identified
+1. [Option A] — [brief description]
+2. [Option B] — [brief description]
+
+## Criteria & Assessment
+| Criterion | Option A | Option B |
+|-----------|----------|----------|
+| [e.g. Complexity] | [assessment] | [assessment] |
+
+## Tradeoff Summary
+- Option A excels at [X] but sacrifices [Y]
+- Option B excels at [Y] but sacrifices [X]
+
+## Recommendation
+[If applicable: "Choose A if [condition]; choose B if [condition]"]
+- **Confidence**: [level and reasoning]
+
+## Risks & Unknowns
+[Uncertainties that could change the recommendation]
+</COUNCIL_MEMBER_RESPONSE>
+\`\`\``,
+
+  PLAN: `
+## Analysis Intent: PLAN
+
+You are conducting a **planning analysis** — your goal is to define current state, target state, and a phased path between them.
+
+**Focus:**
+- Assess where things stand now (current state, constraints, existing assets)
+- Define the target state clearly (what "done" looks like)
+- Break the path into phases with sequencing, dependencies, and exit criteria
+- Identify risks per phase and mitigation strategies
+- Estimate effort where possible (relative sizing is fine: small/medium/large)
+
+**Required output fields:**
+- Current state assessment, Target state definition
+- Phases (each with: goal, tasks, exit criteria, dependencies, risks, effort estimate)
+- Critical path, Key risks and mitigations
+
+**Structure your response as:**
+\`\`\`
+<COUNCIL_MEMBER_RESPONSE>
+## Current State
+[What exists today, constraints, and starting conditions]
+
+## Target State
+[What "done" looks like — measurable where possible]
+
+## Phase 1: [Name]
+- **Goal**: [what this phase achieves]
+- **Tasks**: [concrete work items]
+- **Exit Criteria**: [how you know this phase is complete]
+- **Dependencies**: [what must be true before starting]
+- **Risks**: [what could go wrong, with mitigation]
+- **Effort**: [small/medium/large or time estimate]
+
+## Phase 2: [Name]
+...
+
+## Critical Path & Dependencies
+[Which phases block others, what the minimum viable sequence is]
+
+## Summary
+[Overall effort assessment, key risks, confidence in feasibility]
+</COUNCIL_MEMBER_RESPONSE>
+\`\`\``,
+
+  EXPLAIN: `
+## Analysis Intent: EXPLAIN
+
+You are conducting an **explanatory analysis** — your goal is to build understanding of how something works, why it exists, or what it means.
+
+**Focus:**
+- Lead with a clear thesis statement — your one-sentence answer to the question
+- Identify the key mechanisms, components, or concepts that drive the answer
+- Provide evidence and references for each mechanism (not just assertions)
+- Map relationships, data flows, or causal chains between components
+- Acknowledge unknowns, gaps in evidence, and areas of uncertainty
+- Works for any domain: architecture, external systems, abstract concepts, research topics
+
+**Required output fields:**
+- Thesis statement, Key mechanisms/components (with evidence)
+- Relationships and interactions, Evidence references
+- Unknowns and knowledge gaps, Overall confidence
+
+**Structure your response as:**
+\`\`\`
+<COUNCIL_MEMBER_RESPONSE>
+## Thesis
+[One clear sentence answering the core question]
+
+## Key Mechanisms
+### 1. [Mechanism/Component Name]
+- **What**: [description]
+- **Evidence**: [reference, observation, or reasoning]
+- **Role**: [how this contributes to the overall answer]
+
+### 2. [Mechanism/Component Name]
+...
+
+## Relationships & Interactions
+[How the mechanisms connect — data flows, dependencies, causal chains]
+
+## Unknowns & Gaps
+- [What you could not determine or verify]
+- [Areas where confidence is lower and why]
+
+## Summary
+[Synthesized explanation with overall confidence level]
+</COUNCIL_MEMBER_RESPONSE>
+\`\`\``,
+}

src/agents/athena/council-member-agent.ts

@@ -4,33 +4,26 @@ import { createAgentToolAllowlist } from "../../shared"
 
 const MODE: AgentMode = "subagent"
 
-export const COUNCIL_MEMBER_PROMPT = `You are an independent code analyst in a multi-model analysis council. Your role is to provide thorough, evidence-based analysis.
+export const COUNCIL_MEMBER_PROMPT = `You are an independent analyst in a multi-model analysis council. Your role is to provide thorough, evidence-based analysis.
 
 ## Your Role
 - You are one of several AI models analyzing the same question independently
 - Your analysis should be thorough and evidence-based
 - You are read-only — you cannot modify any files, only analyze
-- Focus on finding real issues, not hypothetical ones
 
 ## Instructions
 1. Analyze the question carefully
-2. Search the codebase thoroughly using available tools (Read, Grep, Glob, LSP)
-3. Report your findings with evidence (file paths, line numbers, code snippets)
-4. For each finding, state:
-   - What the issue/observation is
-   - Where it is (file path, line number)
-   - Why it matters (severity: critical/high/medium/low)
-   - Your confidence level (high/medium/low)
-5. Be concise but thorough — quality over quantity
+2. Use available tools to gather evidence relevant to the question
+3. For each point, state what you observed, where (if applicable), and your confidence level
+4. Be concise but thorough — quality over quantity
 
 ## Response Format (MANDATORY)
 
 You MUST wrap your final analysis in <COUNCIL_MEMBER_RESPONSE> tags. This is how the system extracts your findings.
 
 **Include inside tags:**
-- Key findings with evidence (file paths, line numbers, code snippets)
+- Key findings with supporting evidence
 - Confidence levels for each finding (high/medium/low)
-- Severity assessments (critical/high/medium/low)
 - Concerns and caveats
 
 **Exclude from tags (keep outside):**
@@ -41,14 +34,12 @@ You MUST wrap your final analysis in <COUNCIL_MEMBER_RESPONSE> tags. This is how
 **Example structure:**
 \`\`\`
 <COUNCIL_MEMBER_RESPONSE>
-## Finding 1: [Title]
-- **Location**: src/path/to/file.ts:42-56
-- **Severity**: high
+## Point 1: [Title]
+- **Observation**: [what you found]
 - **Confidence**: high
-- **Issue**: [description]
-- **Evidence**: [code snippet or reference]
+- **Details**: [supporting evidence or reasoning]
 
-## Finding 2: [Title]
+## Point 2: [Title]
 ...
 
 ## Summary

src/hooks/athena-sisyphus-only/agent-matcher.ts

@@ -0,0 +1,5 @@
+import { ATHENA_AGENT } from "./constants"
+
+export function isAthenaAgent(agentName: string | undefined): boolean {
+  return agentName?.toLowerCase().includes(ATHENA_AGENT) ?? false
+}

src/hooks/athena-sisyphus-only/constants.ts

@@ -0,0 +1,5 @@
+export const HOOK_NAME = "athena-sisyphus-only"
+
+export const ATHENA_AGENT = "athena"
+
+export const BLOCKED_TOOLS = ["Write", "Edit", "write", "edit"]

src/hooks/athena-sisyphus-only/hook.ts

@@ -0,0 +1,49 @@
+import type { PluginInput } from "@opencode-ai/plugin"
+import { HOOK_NAME, BLOCKED_TOOLS } from "./constants"
+import { log } from "../../shared/logger"
+import { getAgentFromSession } from "../prometheus-md-only/agent-resolution"
+import { isAthenaAgent } from "./agent-matcher"
+import { isAllowedPath } from "./path-policy"
+
+export function createAthenaSisyphusOnlyHook(ctx: PluginInput) {
+  return {
+    "tool.execute.before": async (
+      input: { tool: string; sessionID: string; callID: string },
+      output: { args: Record<string, unknown>; message?: string }
+    ): Promise<void> => {
+      if (!BLOCKED_TOOLS.includes(input.tool)) {
+        return
+      }
+
+      const agentName = await getAgentFromSession(input.sessionID, ctx.directory, ctx.client)
+
+      if (!isAthenaAgent(agentName)) {
+        return
+      }
+
+      const filePath = (output.args.filePath ?? output.args.path ?? output.args.file) as string | undefined
+      if (!filePath) {
+        return
+      }
+
+      if (!isAllowedPath(filePath, ctx.directory)) {
+        log(`[${HOOK_NAME}] Blocked: Athena attempted write outside .sisyphus/`, {
+          sessionID: input.sessionID,
+          tool: input.tool,
+          filePath,
+          agent: agentName,
+        })
+        throw new Error(
+          `[${HOOK_NAME}] Athena can only write/edit files inside .sisyphus/ directory. Attempted to modify: ${filePath}`
+        )
+      }
+
+      log(`[${HOOK_NAME}] Allowed: .sisyphus/ write permitted`, {
+        sessionID: input.sessionID,
+        tool: input.tool,
+        filePath,
+        agent: agentName,
+      })
+    },
+  }
+}

src/hooks/athena-sisyphus-only/index.ts

@@ -0,0 +1,2 @@
+export * from "./constants"
+export { createAthenaSisyphusOnlyHook } from "./hook"

src/hooks/athena-sisyphus-only/path-policy.ts

@@ -0,0 +1,30 @@
+import { relative, resolve, isAbsolute } from "node:path"
+
+/**
+ * Cross-platform path validator for Athena file writes.
+ * Uses path.resolve/relative instead of string matching to handle:
+ * - Windows backslashes (e.g., .sisyphus\\notepads\\x.yaml)
+ * - Mixed separators (e.g., .sisyphus\\plans/x.md)
+ * - Case-insensitive directory matching
+ * - Workspace confinement (blocks paths outside root or via traversal)
+ * - No extension restriction: any file type is allowed inside .sisyphus/
+ */
+export function isAllowedPath(filePath: string, workspaceRoot: string): boolean {
+  // 1. Resolve to absolute path
+  const resolved = resolve(workspaceRoot, filePath)
+
+  // 2. Get relative path from workspace root
+  const rel = relative(workspaceRoot, resolved)
+
+  // 3. Reject if escapes root (starts with ".." or is absolute)
+  if (rel.startsWith("..") || isAbsolute(rel)) {
+    return false
+  }
+
+  // 4. Check if .sisyphus/ or .sisyphus\ exists anywhere in the path (case-insensitive)
+  if (!/\.sisyphus[/\\]/i.test(rel)) {
+    return false
+  }
+
+  return true
+}

src/shared/agent-tool-restrictions.ts

@@ -48,8 +48,6 @@ const AGENT_RESTRICTIONS: Record<string, Record<string, boolean>> = {
   },
 
   athena: {
-    write: false,
-    edit: false,
     call_omo_agent: false,
   },