feat: add generic critique agent

Makefile

@@ -23,6 +23,7 @@ script-test:
 	$(call run-timed,bash scripts/post-fix-test.sh)
 	$(call run-timed,bash scripts/post-retro-test.sh)
 	$(call run-timed,bash scripts/post-scribe-test.sh)
+	$(call run-timed,bash scripts/post-critique-test.sh)
 	$(call run-timed,bash scripts/validate-output-schema-test.sh)
 
 test: script-test

README.md

@@ -11,6 +11,7 @@ First-class agents for the [fullsend](https://github.com/fullsend-ai/fullsend) p
 | **Review** | Dispatches parallel sub-agents across six review dimensions | PR events, `/fs-review` |
 | **Fix** | Implements targeted fixes from review feedback | Review comments, `/fs-fix` |
 | **Prioritize** | Scores issues using the RICE framework | Schedule, `/fs-prioritize` |
+| **Critique** | Adversarial reviewer and quality gate for refinement plans | `ready-to-critique` label |
 | **Retro** | Analyzes completed workflows and proposes improvements | PR close, `/fs-retro` |
 
 See [`docs/`](docs/) for detailed documentation on each agent.

config.yaml

@@ -9,3 +9,4 @@ agents:
   - source: harness/review.yaml
   - source: harness/retro.yaml
   - source: harness/prioritize.yaml
+  - source: harness/critique.yaml

docs/critique.md

@@ -0,0 +1,71 @@
+# Critique Agent
+
+Adversarial reviewer and quality gate for refinement plans. Evaluates a proposed decomposition from the refine agent and decides: approve (ready for issue creation), revise (send back to refine), or needs_input (escalate to a human for clarification).
+
+> **Note**: The explore and refine agents are added in separate PRs. Cross-references to their docs pages will resolve once all three PRs merge.
+
+## How the agent works
+
+The critique agent runs after the refine agent has produced a decomposition plan. It reads the original work item, exploration context, the proposed plan, and any prior critique history (for revision rounds). It evaluates the plan across seven dimensions — coverage, granularity, dependency coherence, implementability, scope accuracy, assumption grounding, and description clarity — and produces a structured verdict.
+
+The agent runs in a read-only sandbox. It cannot modify issues or create children directly — those mutations are performed by the post-script on the runner. The sandbox policy permits read-only egress to GitHub and Jira APIs plus inference endpoints, but the agent's only output is a structured JSON critique result consumed by the post-script, which posts a summary comment, attaches feedback, and applies labels to signal the next pipeline step.
+
+## How it helps
+
+- Prevents over-decomposition (15 issues when 6 would suffice) and under-decomposition (one giant epic).
+- Catches scope creep — children that exceed what the parent asked for.
+- Validates that acceptance criteria are testable, not vague ("should be fast").
+- Detects assumption laundering — plans that look confident but are built on unverified guesses.
+- Creates child issues automatically on approval when `AUTO_CREATE=true`.
+
+## Commands
+
+| Command | Where | Effect |
+|---------|-------|--------|
+| `/fs-create` | Issue comment | Creates child issues from an approved plan |
+| `/fs-refine` | Issue comment | Restarts the refinement pipeline |
+
+## Pipeline integration
+
+The critique agent is the third stage of a three-stage refinement pipeline:
+
+1. **Explore** — gathers technical context
+2. **Refine** — decomposes the work item
+3. **Critique** — reviews and approves/revises the decomposition (this agent)
+
+### Verdict outcomes
+
+| Verdict | Post-script action |
+|---------|-------------------|
+| `approved` + `AUTO_CREATE=true` | Creates child issues immediately via `create-children.sh` |
+| `approved` + `AUTO_CREATE=false` | Posts approval, adds `refine-approved` label for human gate |
+| `revise` + under round limit | Posts feedback, adds `ready-to-refine` label for revision |
+| `revise` + at round limit | Escalates to human, adds `refine-needs-human` and `refine-escalated` labels |
+| `needs_input` | Posts question, adds `refine-needs-input` label |
+
+## Control labels
+
+These labels are managed by the critique pipeline:
+
+| Label | Meaning |
+|-------|---------|
+| `refine-approved` | Plan approved; ready for human review or auto-creation |
+| `ready-to-refine` | Revisions requested; refine agent should re-run |
+| `refine-revision-round-N` | Tracks which revision round the pipeline is on |
+| `refine-needs-input` | Human must answer a question before proceeding |
+| `refine-needs-human` | Max review rounds reached; human decision needed |
+| `refine-escalated` | Distinguishes max-rounds escalation from genuine approval |
+
+## Child issue creation
+
+When the critique agent approves a plan and `AUTO_CREATE=true`, the post-script sources `create-children.sh` to create child issues. This script:
+
+- Creates issues in topological order (parents before children)
+- Supports both GitHub (sub-issues API) and Jira (parent-child hierarchy with fallback to "Relates" links)
+- Deduplicates against existing children to prevent double-creation on re-runs
+- Resolves Jira issue types against the project's available types
+- Handles cross-project creation when a routing skill provides `target_project`
+
+## Source
+
+[`harness/critique.yaml`](../harness/critique.yaml)

env/critique.env

@@ -0,0 +1,6 @@
+export ISSUE_CONTEXT=/sandbox/workspace/issue-context.json
+export EXPLORE_CONTEXT=/sandbox/workspace/exploration_context.json
+export REFINE_RESULT=/sandbox/workspace/refine-result.json
+export CRITIQUE_HISTORY=/sandbox/workspace/critique-history.json
+export PROJECT_ROUTING=/sandbox/workspace/routing-skill.md
+export PLATFORM_CONTEXT=/sandbox/workspace/platform-context.md

harness/critique.yaml

@@ -0,0 +1,73 @@
+---
+agent: agents/critique.md
+doc: docs/critique.md
+model: opus
+image: ghcr.io/fullsend-ai/fullsend-sandbox:latest
+policy: policies/critique.yaml
+
+role: critique
+slug: fullsend-ai-critique
+
+host_files:
+  - src: common/env/gcp-vertex.env
+    dest: /sandbox/workspace/.env.d/gcp-vertex.env
+    expand: true
+  - src: env/critique.env
+    dest: /sandbox/workspace/.env.d/critique.env
+    expand: true
+  - src: ${GOOGLE_APPLICATION_CREDENTIALS}
+    dest: /tmp/.gcp-credentials.json
+  - src: ${GCP_OIDC_TOKEN_FILE}
+    dest: /sandbox/workspace/.gcp-oidc-token
+    optional: true
+  - src: /tmp/workspace/issue-context.json
+    dest: /sandbox/workspace/issue-context.json
+    optional: true
+  - src: /tmp/workspace/exploration_context.json
+    dest: /sandbox/workspace/exploration_context.json
+    optional: true
+  - src: /tmp/workspace/refine-result.json
+    dest: /sandbox/workspace/refine-result.json
+    optional: true
+  - src: /tmp/workspace/critique-history.json
+    dest: /sandbox/workspace/critique-history.json
+    optional: true
+  - src: /tmp/workspace/routing-skill.md
+    dest: /sandbox/workspace/routing-skill.md
+    optional: true
+  - src: /tmp/workspace/platform-context.md
+    dest: /sandbox/workspace/platform-context.md
+    optional: true
+
+pre_script: scripts/pre-critique.sh
+post_script: scripts/post-critique.sh
+
+validation_loop:
+  script: scripts/validate-output-schema.sh
+  schema: schemas/critique-result.schema.json
+  max_iterations: 2
+
+timeout_minutes: 15
+
+env:
+  runner:
+    ISSUE_KEY: "${ISSUE_KEY}"
+    ISSUE_SOURCE: "${ISSUE_SOURCE}"
+    REPO_FULL_NAME: "${REPO_FULL_NAME}"
+    REFINE_RUN_ID: "${REFINE_RUN_ID}"
+    GITHUB_ISSUE_NUMBER: "${GITHUB_ISSUE_NUMBER}"
+    REVIEW_ROUND: "${REVIEW_ROUND}"
+    MAX_REVIEW_ROUNDS: "${MAX_REVIEW_ROUNDS}"
+    AUTO_CREATE: "${AUTO_CREATE}"
+  sandbox:
+    REVIEW_ROUND: "${REVIEW_ROUND}"
+    MAX_REVIEW_ROUNDS: "${MAX_REVIEW_ROUNDS}"
+
+forge:
+  github:
+    pre_script: scripts/pre-critique.sh
+    post_script: scripts/post-critique.sh
+    env:
+      runner:
+        GH_TOKEN: "${GH_TOKEN}"
+        PUSH_TOKEN: "${PUSH_TOKEN}"

policies/critique.yaml

@@ -0,0 +1,71 @@
+---
+version: 1
+
+# Sandbox policy for the critique agent.
+#
+# Read-only agent: needs GitHub API for issue context and Jira for
+# attachment retrieval. No write access — the post-script handles
+# mutations on the runner.
+
+filesystem_policy:
+  include_workdir: true
+  read_only: [/usr, /lib, /proc, /dev/urandom, /app, /etc, /var/log]
+  read_write: [/sandbox, /tmp, /dev/null]
+landlock:
+  compatibility: best_effort
+process:
+  run_as_user: sandbox
+  run_as_group: sandbox
+
+network_policies:
+  vertex_ai:
+    name: vertex-ai
+    endpoints:
+      - host: "api.anthropic.com"
+        port: 443
+        protocol: rest
+        enforcement: enforce
+        access: read-write
+      - host: "*.googleapis.com"
+        port: 443
+        protocol: rest
+        enforcement: enforce
+        access: read-write
+    binaries:
+      - path: "**/claude"
+      - path: "**/node"
+
+  github_api:
+    name: github-api
+    endpoints:
+      - host: "api.github.com"
+        port: 443
+        protocol: rest
+        enforcement: enforce
+        access: read-only
+      - host: "github.com"
+        port: 443
+        protocol: rest
+        enforcement: enforce
+        access: read-only
+      - host: "raw.githubusercontent.com"
+        port: 443
+        protocol: rest
+        enforcement: enforce
+        access: read-only
+    binaries:
+      - path: "**/gh"
+      - path: "**/git"
+      - path: "**/node"
+
+  jira_api:
+    name: jira-api
+    endpoints:
+      - host: "*.atlassian.net"
+        port: 443
+        protocol: rest
+        enforcement: enforce
+        access: read-only
+    binaries:
+      - path: "**/curl"
+      - path: "**/node"