Create Spike Agent Skill

Create Spike

Investigate a problem, map it to the codebase, and produce a structured GitHub issue ready for human disposition and roadmap placement.

A spike is an exploratory investigation. The user has a vague idea — a feature they want, a bug they've noticed, a performance concern — but hasn't mapped it to code, assessed feasibility, or structured it as a buildable issue. This skill does that mapping.

Prerequisites

• The gh CLI must be authenticated (gh auth status)
• You must be in a git repository with a GitHub remote

Workflow Overview

User describes a problem
  │
  ├─ Step 1: Gather the problem statement
  │   └─ Ask ONE round of clarifying questions if genuinely needed
  │
  ├─ Step 2: Deep codebase investigation via principal-engineer-reviewer
  │   └─ Map the problem to code, assess feasibility, identify risks
  │
  ├─ Step 3: Determine labels from the repo
  │
  ├─ Step 4: Create a GitHub issue with structured findings
  │
  └─ Step 5: Report to user with issue URL and next steps

Step 1: Gather the Problem Statement

The user provides a problem description. This could be:

• A feature idea: "I want sandboxes to be able to reach private IPs"
• A bug report: "The retry logic in the proxy seems too aggressive"
• A performance concern: "Policy evaluation is slow for large rule sets"
• A refactoring goal: "The config parsing is scattered across too many modules"

Extract from the user's input:

• What they want (the desired outcome or observed problem)
• Why they want it (motivation, use case, or trigger)
• Constraints they've mentioned (backwards compatibility, performance targets, etc.)

Clarification policy

If the problem is too vague to determine which area of the codebase to investigate, ask ONE round of clarifying questions. Do not over-interrogate. Examples of when to ask:

• "Make things faster" — ask which component or operation is slow
• "Fix the networking" — ask what specific behavior is wrong

Examples of when NOT to ask:

• "The retry logic in the proxy is too aggressive" — clear enough, start investigating
• "Allow sandbox egress to private IP space" — clear enough, start investigating
• "The OPA policy evaluation needs caching" — clear enough, start investigating

Step 2: Deep Codebase Investigation

This is the core of the skill. Use the Task tool with the principal-engineer-reviewer sub-agent to perform a thorough codebase investigation.

Task tool with subagent_type="principal-engineer-reviewer"

The prompt to the reviewer must instruct it to:

• Identify which components/subsystems are involved. Don't just guess from names — read the code to confirm.

• Read the relevant source files thoroughly. Not just grep for keywords — actually read and understand the logic. Follow the call chain from entry point through to the relevant behavior.

• Map the current architecture for the affected area. How do the components interact? What's the data flow? Where are the boundaries?

• Identify the exact code paths that would need to change. Provide file paths and line numbers. Name the functions, structs, and modules.

• Assess feasibility and complexity:
• Low: Isolated change, < 3 files, clear path forward
• Medium: Multiple files/components, some design decisions, but well-scoped
• High: Cross-cutting changes, architectural decisions needed, significant unknowns

• Identify risks, edge cases, and design decisions that need human input. What could go wrong? What trade-offs exist? What decisions shouldn't be made by an agent?

• Check for existing patterns in the codebase that should be followed. If there's a convention for how similar features are implemented, note it. The implementation should be consistent.

• Look at relevant tests to understand test coverage expectations. What test patterns exist? What level of coverage is expected for this area?

• Check architecture docs in the architecture/ directory for relevant documentation about the affected subsystems.

10. Assess gateway config documentation impact. If the change would add, remove, rename, or change defaults for gateway TOML keys or driver-specific config options, call out that docs/reference/gateway-config.mdx must be updated. If the change is surfaced through Helm or compute-driver setup docs, call out the relevant deployment or compute-driver docs too.

11. Assess Linux Security Module (LSM) impact. If the change involves process identity, /proc filesystem access, file labeling, binary execution, or inter-process visibility, call out whether it will behave differently on hosts running SELinux (enforcing) or AppArmor. For example: reading /proc/<pid>/exe across an SELinux domain boundary returns ENOENT, not EACCES. Tests that fork+exec into system binaries (different SELinux label) will fail on enforcing hosts. Flag any LSM-sensitive code paths and recommend mitigations.

12. Determine the issue type: feat, fix, refactor, chore, perf, or docs.

What makes a good investigation prompt

Include in the prompt to the reviewer:

• The user's problem statement (verbatim or lightly paraphrased)
• Any constraints the user mentioned
• A clear instruction to return: component list, file references with line numbers, architecture summary, feasibility assessment, risks, and the issue type

What to do with the results

The reviewer will return a detailed analysis. You'll use this to populate the issue body (Step 4). The issue should contain both the stakeholder-readable summary and the full technical investigation — everything in one place.

Step 3: Determine Labels

Fetch the available labels from the repository:

gh label list --limit 100

Based on the investigation results, select appropriate labels:

• Do not add issue type labels — GitHub built-in issue types come from issue templates or manual follow-up, not labels
• Include area labels if they exist in the repo (e.g., area:sandbox, area:proxy, area:policy, area:cli)
• Do not invent labels — only use labels that already exist in the repo
• Add state:validated only when the evidence is sufficient for human disposition — the spike established a coherent problem or proposal and completed the factual assessment needed for a human yes/no decision
• Add state:needs-info instead when material evidence is missing — identify the exact evidence, reproduction details, or decision input still needed in the issue body
• Never add state:accepted, an agent:* label, or the roadmap label — acceptance, roadmap placement, and requests for agent work require a human decision

Step 4: Create the GitHub Issue

Create the issue with a structured body containing both the stakeholder-readable summary and the full technical investigation. The title should follow conventional commit format.

gh issue create \
  --title "<type>: <concise description of the problem/feature>" \
  --label "<area:component>" --label "<state:validated|state:needs-info>" \
  --body "$(cat <<'EOF'
## Problem Statement

<What and why — refined from the user's description. 2-4 sentences. Written for stakeholders, not just engineers.>

## Technical Context

<What the investigation found about the current architecture in the affected area. How things work today and why a change is needed.>

## Affected Components

| Component | Key Files | Role |
|-----------|-----------|------|
| <component> | `<file1>`, `<file2>` | <what this component does in the context of this change> |
| ... | ... | ... |

## Technical Investigation

### Architecture Overview

<How the affected subsystems work today. Include data flow, component interactions, and relevant design decisions. Reference architecture docs if applicable.>

### Code References

| Location | Description |
|----------|-------------|
| `<file>:<line>` | <what this code does and why it's relevant> |
| `<file>:<line>` | <what this code does and why it's relevant> |
| ... | ... |

### Current Behavior

<What happens today in the code paths that would change. Be specific — name functions, trace the flow.>

### What Would Need to Change

<Detailed breakdown of modifications needed, organized by component. Include specific functions and structs, but stop short of writing an implementation plan — that's `build-from-issue`'s job.>

### Alternative Approaches Considered

<If the investigation surfaced multiple viable approaches, describe them and note trade-offs. Flag which decisions need human input.>

### Patterns to Follow

<Existing patterns in the codebase that the implementation should be consistent with. Reference specific examples.>

## Proposed Approach

<High-level strategy — NOT a full implementation plan. That's `build-from-issue`'s job. Describe the direction, not the steps. 3-6 sentences.>

## Scope Assessment

- **Complexity:** <Low / Medium / High>
- **Confidence:** <High — clear path / Medium — some unknowns / Low — needs discussion>
- **Estimated files to change:** <count>
- **Issue type:** `<feat|fix|refactor|chore|perf|docs>`

## Risks & Open Questions

- <risk or unknown that needs human judgment>
- <design decision that could go either way>
- ...

## Disposition Readiness

- **State:** `<state:validated|state:needs-info>`
- **Assessment:** <why the available evidence is or is not sufficient for a human accept/decline decision>
- **Missing evidence:** <specific evidence still needed, or "None">

## Test Considerations

- <what testing strategy makes sense for this change>
- <which test levels are needed: unit, integration, e2e>
- <any test infrastructure that may need to be added>
- <what tests exist for the affected area today, what patterns should be followed, any test infrastructure gaps>

---
*Created by spike investigation. `state:validated` means the issue is ready for human disposition; `state:needs-info` means specific evidence is still required. A human applies `state:accepted` if OpenShell should pursue the work and places it on the roadmap separately. To queue unattended agent planning, a human applies `agent:plan-requested`; a direct request to an agent does not require that label.*
EOF
)"

Do NOT post a follow-up comment on the issue. All findings must be contained in the issue body itself.

Display the issue URL so it's easily clickable:

Created issue [#<number>](https://github.com/OWNER/REPO/issues/<number>)

Step 5: Report to User

After creating the issue, report:

• The issue URL (as a clickable markdown link)
• A 2-3 sentence summary of what was found
• Key risks or decisions that need human attention
• Next steps:

For state:validated:

> Review the issue and decide whether OpenShell should pursue it. If yes, replace state:validated with state:accepted and separately associate it with a roadmap item. The work may remain human-owned. Apply agent:plan-requested to queue planning for an unattended agent, or directly ask an agent to use build-from-issue. If no, close it as not planned and record the rationale.

For state:needs-info:

> Collect the missing evidence identified in the issue. Leave it off the roadmap. Once the evidence is sufficient, replace state:needs-info with state:validated for human disposition.

Design Principles

• Everything goes in the issue body. Do NOT post follow-up comments. The issue body should contain both the stakeholder-readable summary and the full technical investigation, all in one place.

• Do NOT create an implementation plan. The spike identifies the problem space and proposes a direction. The implementation plan is build-from-issue's responsibility, created after human review of the spike.

• One round of clarification max. Don't turn this into an interrogation. If the user provides enough to identify the area of the codebase, start investigating.

• The issue should save build-from-issue work. When build-from-issue runs, it reads the issue body as input context. The technical investigation section should contain enough detail that its principal-engineer-reviewer can build on the investigation rather than starting from scratch.

• Cross-reference build-from-issue. Mention it as the natural next step in the issue body footer.

• Treat validation as an evidence threshold, not an automatic spike outcome. Apply state:validated only when the investigation supports a human accept/decline decision. Otherwise apply state:needs-info, state what is missing, and leave the issue off the roadmap.

Useful Commands Reference

| Command | Description |

| --- | --- |

| gh issue create --title "..." --body "..." --label "..." | Create a new issue |

| gh label list --limit 100 | List available labels in the repo |

| gh issue edit <id> --add-label "..." | Add labels to an issue |

| gh issue view <id> --json number,title,body,state,labels | Fetch issue metadata |

Example Usage

Feature spike

User says: "Allow sandbox egress to private IP space via networking policy"

• Problem is clear — no clarification needed
• Fire principal-engineer-reviewer to investigate:
• Finds is_internal_ip() SSRF check in proxy.rs that blocks RFC 1918 addresses
• Reads OPA policy evaluation pipeline in opa.rs and crates/openshell-sandbox/data/sandbox-policy.rego
• Reads proto definitions in sandbox.proto for NetworkEndpoint
• Maps the 4-layer defense model: netns, seccomp, OPA, SSRF check
• Reads architecture/security-policy.md and architecture/sandbox.md
• Identifies exact insertion points: policy field addition, SSRF check bypass path, OPA rule extension
• Assesses: Medium complexity, High confidence, ~6 files
• Fetch labels — select area:sandbox, area:proxy, area:policy, state:validated
• Create issue: feat: allow sandbox egress to private IP space via networking policy — body includes both the summary and full investigation (code references, architecture context, alternative approaches)
• Report: "Created issue #59. The investigation found that private IP blocking is enforced at the SSRF check layer in the proxy. The proposed approach adds a policy-level override. A human must now accept or decline it and place it on the roadmap if accepted."

Bug investigation spike

User says: "The proxy retry logic seems too aggressive — I'm seeing cascading failures under load"

• Problem is clear enough — investigate retry behavior in the proxy
• Fire principal-engineer-reviewer:
• Finds retry configuration in proxy request handling
• Reads the retry loop, backoff strategy, and timeout settings
• Checks if there's circuit breaker logic
• Maps the failure propagation path
• Identifies that retries happen without backoff jitter, causing thundering herd
• Assesses: Low complexity, High confidence, ~2 files
• Fetch labels — select area:proxy, state:validated
• Create issue: fix: proxy retry logic causes cascading failures under load — body includes both the summary and full investigation (retry code references, current behavior trace, comparison to standard backoff patterns)
• Report: "Created issue #74. The proxy retries without jitter or circuit breaking, which amplifies failures under load. A human must now accept or decline it and place it on the roadmap if accepted."

Performance/refactoring spike

User says: "Policy evaluation is getting slow — can we cache compiled OPA policies?"

• Problem is clear — investigate OPA policy evaluation performance
• Fire principal-engineer-reviewer:
• Reads the OPA evaluation pipeline end to end
• Measures where policies are loaded and compiled (per-request vs. cached)
• Checks if there's an existing caching layer
• Reads the policy reload/hot-swap mechanism
• Identifies that policies are recompiled on every evaluation
• Assesses: Medium complexity, Medium confidence (cache invalidation is a design decision), ~4 files
• Fetch labels — select area:policy, state:validated
• Create issue: perf: cache compiled OPA policies to reduce evaluation latency — body includes both the summary and full investigation (compilation hot path, per-request overhead, cache invalidation strategies with trade-offs)
• Report: "Created issue #81. Policies are recompiled per-request with no caching. The main design decision is the cache invalidation strategy. A human must now accept or decline it and place it on the roadmap if accepted."