Challenge Skill Skill | Agent Skills

Challenge Skill

Purpose

Dispatch the challenger agent as a dedicated subagent for operational feasibility scrutiny. This is a mandatory workflow step for both oneoff-spec and orchestrator workflows.

pre-implementation and pre-orchestration: Run as a convergence loop (2 consecutive clean passes, auto-decision engine integration)
pre-test and pre-review: Run as a single-pass check

The embedded pre-flight questions in each skill still run as part of normal skill execution -- this dedicated dispatch provides additional scrutiny as a separate workflow step.

Usage

/challenge <spec-group-id> --stage <stage>

Stage values: pre-implementation, pre-test, pre-review, pre-orchestration

When to Use

The dedicated /challenge dispatch is MANDATORY at all 4 stages for oneoff-spec and orchestrator workflows:

pre-implementation (convergence loop): After investigation convergence, before implementation begins. Fix agent: implementer.
pre-test (single-pass): After implementation completes, before test verification gates (Integration Verify). Validates test fixtures, test data availability, test infrastructure readiness.
pre-review (single-pass): After Unify (loop), before Code Review (loop). Identifies riskiest change areas, integration surfaces crossed, review focus recommendations.
pre-orchestration (convergence loop, orchestrator only): After investigation convergence, before /orchestrate begins. Fix agent: spec-author.

The dedicated /challenge dispatch is NOT required for:

oneoff-vibe workflows: Too lightweight to warrant dedicated scrutiny
refactor workflows: Behavior preservation is enforced by tests, not pre-flight checks
journal-only workflows: No implementation to challenge

Process

Step 1: Determine Stage and Mode

| Workflow Phase | Stage Parameter | Mode | Fix Agent | | --------------------- | -------------------- | ---------------- | ------------- | | Before implementation | pre-implementation | Convergence loop | implementer | | Before test writing | pre-test | Single-pass | N/A | | Before code review | pre-review | Single-pass | N/A | | Before orchestration | pre-orchestration | Convergence loop | spec-author |

Step 2: Gather Stage-Specific Input Context

| Stage | Required Input Context | | -------------------- | ------------------------------------------------------------------------------------ | | pre-implementation | Approved spec, environment configuration, dependency manifest, execution environment | | pre-test | Approved spec, implementation artifacts (file paths), test infrastructure inventory | | pre-review | Spec, implementation diff/artifacts, integration boundary list | | pre-orchestration | MasterSpec/WorkstreamSpecs, workstream dependency graph, shared resource inventory |

Step 3: Execute (Convergence Loop or Single-Pass)

For convergence loop stages (pre-implementation, pre-orchestration):

Loop state (owned by this skill):

{
  "gate": "challenger",
  "iteration_count": 0,
  "clean_pass_count": 0,
  "max_iterations": 5,
  "required_clean_passes": 2,
  "findings_history": [],
  "cross_stage_resolution_count": 0,
  "cross_stage_resolution_cap": 3
}

Loop mechanics:

Dispatch challenger for one pass:

Task: challenger
Prompt: |
  Stage: <stage parameter>
  Spec group: <spec-group-id>
  Pass: <iteration_count + 1>

  Input context:
    <stage-specific context from Step 2>

  Perform deep operational feasibility scrutiny.
  Produce severity-rated findings (Critical/High/Medium/Low) with:
    - Structured confidence enum (high/medium/low)
    - Deterministic finding IDs in format chk-{category}-{hash}
  Reference env var names only -- never actual secret values.
  Return < 200 words.

Evaluate findings: If no Medium+ findings, increment clean_pass_count. Otherwise reset to 0.
Auto-decision engine: For Medium+ findings, invoke auto-decision engine:
- Findings with valid recommendations (action verb + field reference + high/medium confidence) are auto-accepted
- Security-tagged, low-confidence, and ambiguous findings escalate to human
- Oscillation detection: if a finding ID recurs after its fix was applied, escalate immediately
- All-or-nothing batch processing: if engine crashes, present all findings to human
Apply fixes: Dispatch fix agent (implementer or spec-author) with accepted and resolved findings.
Check convergence:
- If clean_pass_count >= 2: Converged. Record convergence (see Step 4 below).
- If iteration_count >= 5: Escalate to human with iteration history.
- Otherwise: Back to step 1.

Cross-stage resolution cap: If a fix at one stage introduces a blocker at another stage (e.g., fixing a pre-implementation finding creates a pre-test conflict), increment a cross-stage counter. When the count reaches 3, escalate to the human with the full blocker chain. This prevents infinite oscillation between stages.

For single-pass stages (pre-test, pre-review):

Dispatch challenger once and process findings:

Task: challenger
Prompt: |
  Stage: <stage parameter>
  Spec group: <spec-group-id>

  Input context:
    <stage-specific context from Step 2>

  Perform deep operational feasibility scrutiny.
  Produce severity-rated findings (Critical/High/Medium/Low).
  Reference env var names only -- never actual secret values.
  Return < 200 words.

Process findings directly:

Critical/High findings: Block -- resolve before proceeding to the target stage
Medium findings: Log as warnings, proceed with awareness
Low findings: Log only

Step 4: Record Convergence (Convergence Loop Stages Only)

After 2 consecutive clean passes:

# Set manifest flat boolean for PHASE_OBLIGATIONS
node -e "
const fs = require('fs');
const path = '<spec-group-dir>/manifest.json';
const m = JSON.parse(fs.readFileSync(path));
m.convergence = m.convergence || {};
m.convergence.challenger_converged = true;
fs.writeFileSync(path, JSON.stringify(m, null, 2) + '\\n');
"

# Set session.json for coercive enforcement
node .claude/scripts/session-checkpoint.mjs update-convergence challenger

Finding Deduplication

If both /challenge and /investigate produce findings about the same issue:

Investigation findings take precedence (formal convergence gate vs. challenger)
Deduplication occurs at the orchestrator level

Integration with Workflow

Embedded pre-flight (always active in each skill):
  Subagent reads SKILL.md -> Encounters Pre-Flight Challenge section -> Addresses questions inline

Dedicated dispatch (MANDATORY at all 4 stages for oneoff-spec and orchestrator, this skill):
  pre-implementation: After investigation convergence -> /challenge (convergence loop) -> Auto-Approval -> Implementation begins
  pre-test:           After impl   -> /challenge --stage pre-test (single-pass) -> Proceed or block -> Integration Verify begins
  pre-review:         After unify  -> /challenge --stage pre-review (single-pass) -> Proceed or block -> Code Review begins
  pre-orchestration:  After investigation convergence -> /challenge (convergence loop) -> Auto-Approval -> /orchestrate begins

Examples

Example 1: Pre-Implementation Challenge (Convergence Loop)

/challenge sg-auth-system --stage pre-implementation

Iteration 1:
  Findings: 1 High (auto-accepted), 0 Critical
  Auto-accepted: chk-env-a1b2c3d4 (add DATABASE_URL to .env)
  Fix agent (implementer) applied recommendation.

Iteration 2: Clean pass (clean_pass_count = 1)
Iteration 3: Clean pass (clean_pass_count = 2)

Convergence achieved in 3 iterations.
challenger_converged = true recorded in manifest.

Example 2: Pre-Orchestration Challenge (Convergence Loop)

/challenge ms-deployment-pipeline --stage pre-orchestration

Iteration 1:
  Findings: 0 Critical, 0 High, 2 Medium
  Auto-accepted: chk-resource-b2c3d4e5 (document shared CONTAINER_REGISTRY)
  Escalated: chk-health-c3d4e5f6 (no health check endpoint -- security-tagged)
  Human resolved: accepted health check recommendation

Iteration 2: Clean pass (clean_pass_count = 1)
Iteration 3: Clean pass (clean_pass_count = 2)

Convergence achieved in 3 iterations.

Example 3: Pre-Test Challenge (Single-Pass)

/challenge sg-auth-system --stage pre-test

Challenger findings:
  Critical: 0
  High: 0
  Medium: 1
    CHK-001: Test database not seeded with required fixtures
  Low: 1

Action: Log warnings, proceed to Integration Verify.

Agent Skills: Challenge Skill

Install this agent skill to your local

Skill Files