Skip to content
Back to skills

Dispatching Code Review

ASecurity

Dispatch a fresh-context code review subagent.

  • 3 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 16, 2026
ai-agentsgosqlcode-reviewgitdatabasesecurity

Works with

  • terminal
  • cli

Security analysis

A100/100

Scanned November 16, 2026

npx -y skills add Acendas/shipyard --skill dispatching-code-review --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dispatching Code Review?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Dispatching Code Review
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/acendas-dispatching-code-review/badge)](https://www.skillsdirectory.com/skills/acendas-dispatching-code-review)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: dispatching-code-review
description: Dispatch a fresh-context code review subagent.
disable-model-invocation: true
---

# Dispatching a Code Review

**Render before asking.** Render the decision context as chat text before any AskUserQuestion (rules printed at startup under SHIPYARD_STARTUP_RULES).

Companion to `dispatching-spec-review` (spec review: "is what was asked delivered?"; code review: "is it any good?").

## When to Invoke

| Caller | Scope | Trigger |
|---|---|---|
| `/ship-execute` post-task gate | One task | Optional — fires only when `effort: M\|L\|XL`; effort: S skips both spec and code review |
| `/ship-execute` wave VERIFY | Wave-level diff | Optional — fires when wave touched security-relevant, financial-domain, **or database/persistence** code (migrations, schema, queries, repositories) |
| `/ship-review` | Sprint or feature | **Required** before user approval |
| `/ship-execute --task` (land-now) | Single-change diff | Optional flag (`--review`) |

## Inputs

- `scope` — `"task" | "wave" | "feature" | "sprint"`
- `target_ids` — list of task / feature IDs (or null when scope is sprint).
- `base_ref` / `head_ref` — diff range.
- `concerns` — subset of `["security", "bugs", "silent-failures", "patterns", "tests", "observability", "data"]`. Default: all. The **`data`** concern **auto-gates** — it produces findings only when the diff touches persistence (migrations, schema, SQL/ORM, repositories, indexes) and no-ops otherwise, so including it by default is free. Caller can narrow (e.g., `["security", "bugs"]` for a wave that didn't touch tests), but **whenever the diff touches the database, `data` MUST be among the concerns** (it is, under the default). In parallel-split dispatch (see `ship-review/references/code-review-orchestration.md`), assign `data` to exactly one subagent so it isn't dropped or duplicated.
- `data_dir` — literal `<SHIPYARD_DATA>` path.
- `project_rules_path` — `.claude/rules/*.md` paths so the patterns scanner has the project's conventions. Shipyard does not inject its own rules into `.claude/rules/`; only project-authored rules pass through here.
- `quality_standards_path` — the literal path to `project-files/references/code-quality-standards.md`. Always included (the caller already effort-gates), and not narrowed by `concerns`.

## Dispatching the Reviewer

The reviewer methodology lives in the registered agent `agents/shipyard-code-reviewer.md`; do not re-inline it here.

**Model + effort (resolved by CLI).** Run `shipyard-data resolve-model think` — it prints one line `model=<m> effort=<e>` from config.md (an empty value means OMIT that field on the Agent call so the subagent inherits; `inherit` is already normalized to empty). Never derive either from config.md prose and never hardcode a model literal. Applies to every dispatch, including each subagent of the parallel-split variant below. Then log it: `shipyard-data events emit dispatch_model_resolved role=think model=<m or inherit> effort=<e or inherit> skill=<this skill>` — the wrapper logs, the resolver stays silent, and a logging failure never blocks the dispatch.

**Plugin-relative paths are resolved here, not in the agent.** `${CLAUDE_PLUGIN_ROOT}` is not verified to expand inside a registered agent's body — resolve the data-implementation guide path (when the `data` concern is gated in) and `quality_standards_path` (always) to literal paths before including them in the brief.

Dispatch:

```
Agent(
  subagent_type: "shipyard:shipyard-code-reviewer",
  model: <model= from resolve-model, or omit if empty>,
  effort: <effort= from resolve-model, or omit if empty>,
  prompt: "
    Scope:              {{scope}}
    Target IDs:          {{target_ids}}
    Base ref:            {{base_ref}}
    Head ref:            {{head_ref}}
    Concerns:            {{concerns_csv}}
    Data dir:            {{data_dir}}
    Project rules:       {{project_rules_path}}
    Quality standards:   {{quality_standards_path}}
    {{data_impl_guide path, if `data` concern is gated in — otherwise omit}}
  "
)
```

## Orchestrator-Side Action Rules

The reviewer's return always carries `STATUS:`, `SCOPE:`, and `TARGETS:`; a `FINDINGS:` count (0 on CLEAN, an integer on FINDINGS) and, on FINDINGS, an `ADVISORY:` count for the 60–80 band, accompany the per-finding block.

1. **`STATUS: CLEAN`** → record; advance.

2. **`STATUS: FINDINGS`**:

   - **High-confidence security findings** (`concern: security`, confidence ≥ 90) → block. Re-dispatch `dispatching-task-loop` with the security findings inlined: *"Code review found security issues that must be fixed: <list>; re-implement and re-probe."*
   - **Other ≥ 80 findings in `/ship-review`** → auto-fix by default through the Stage 0 code-review loop. Do not ask the user about routine fixes; render findings only at user gates or terminal escalation. AskUserQuestion is reserved for severe/risky cases (product decision, destructive migration, credential/security-policy choice, large dependency/platform change, accepting a known defect).
   - **Other ≥ 80 findings outside `/ship-review`** → present in the calling skill's report. 'Present' means the full finding blocks (file:line, snippet, reason) rendered as chat text before any per-finding decision ask — findings that exist only in this skill's return are not presented. The post-task path can auto-redispatch the task once for high-density findings (≥ 3) but stops there to avoid loop-on-quality.
   - **Advisory (60–80)** → log to PROGRESS.md deviations; no auto-action.

3. **`STATUS: BLOCKED`** → quote the subagent's `REASON:` paragraph verbatim as chat text (it exists only in the Agent return — content in a subagent return or in the question/option strings does not count as shown), then AskUserQuestion.

4. **Silent return** — the Agent return is present but no `STATUS:` line appears, or the body is empty/whitespace. Treat this as its own outcome, distinct from CLEAN/FINDINGS/BLOCKED: emit `shipyard-data events emit code_review_dispatch_returned scope=<scope> targets=<target_ids> status=needs-attention reason=silent_return` and re-dispatch ONCE with the same brief. If the re-dispatch is also silent, stop re-dispatching and surface it as a `STATUS: BLOCKED`-shaped ask instead of looping.

5. **Read-only enforcement** — as in `dispatching-spec-review`: post-return `git status --porcelain` + HEAD ref check; any drift is a contract violation → `STATUS: BLOCKED`.

## Parallel Dispatch For High-Stakes Reviews

For high-stakes reviews (release-bound, large diff, payments/auth/data), `/ship-review` may dispatch this skill multiple times in parallel with non-overlapping `concerns` arrays (more tokens, more depth per concern). Assign the `data` concern to exactly one split (see Inputs above) — it must never be dropped or duplicated across splits.

## Pairing With Other Skills

- **`dispatching-spec-review`** runs before this — spec compliance is "did we deliver"; code review is "is the delivery any good." Both must pass for `/ship-review` approval.
- **`dispatching-task-loop`** is re-dispatched when high-confidence findings demand re-implementation.
- **`anti-stub-scan`** is structural; this skill is semantic. **`running-acceptance-probe`** is orthogonal — code review doesn't run probes.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.