gsd-eval-review

name	gsd-eval-review
description	Audit an executed AI phase's evaluation coverage and produce an EVAL-REVIEW.md remediation plan.
metadata	{"short-description":"Audit an executed AI phase's evaluation coverage and produce an EVAL-REVIEW.md remediation plan."}

<codex_skill_adapter>

A. Skill Invocation

• This skill is invoked by mentioning $gsd-eval-review.
• Treat all user text after $gsd-eval-review as {{GSD_ARGS}}.
• If no arguments are present, treat {{GSD_ARGS}} as empty.

B. AskUserQuestion → request_user_input Mapping

GSD workflows use AskUserQuestion (Claude Code syntax). Translate to Codex request_user_input:

Parameter mapping:

• header → header
• question → question
• Options formatted as "Label" — description → {label: "Label", description: "description"}
• Generate id from header: lowercase, replace spaces with underscores

Batched calls:

• AskUserQuestion([q1, q2]) → single request_user_input with multiple entries in questions[]

Multi-select workaround:

• Codex has no multiSelect. Use sequential single-selects, or present a numbered freeform list asking the user to enter comma-separated numbers.

Execute mode fallback:

• When request_user_input is rejected or unavailable, activate TEXT_MODE: append --text to {{GSD_ARGS}} so the workflow's built-in text-mode branching takes over. Present every AskUserQuestion call as a plain-text numbered list, then stop and wait for the user's reply. Do NOT pick a default and continue (#3018 / #3808).
• You may only proceed without a user answer when one of these is true: (a) the invocation included an explicit non-interactive flag (--auto or --all), (b) the user has explicitly approved a specific default for this question, or (c) the workflow's documented contract says defaults are safe (e.g. autonomous lifecycle paths).
• Do NOT write workflow artifacts (CONTEXT.md, DISCUSSION-LOG.md, PLAN.md, checkpoint files) until the user has answered the plain-text questions or one of (a)-(c) above applies. Surfacing the questions and waiting is the correct response — silently defaulting and writing artifacts is the #3018 failure mode.

C. Task() → spawn_agent Mapping

GSD workflows use Task(...) (Claude Code syntax). Translate to Codex collaboration tools:

Schema detection (required first step): Codex exposes two spawn_agent schemas:

• agent_type-capable schema (e.g. multi_agent_v2): spawn_agent accepts agent_type, message, reasoning_effort, fork_context, etc. — typed GSD agent dispatch is available.
• Generic schema (multi_agent_v1): spawn_agent accepts only message, items, fork_context — there is no agent_type field. Typed GSD agent dispatch is unavailable in this session.

Before spawning, inspect the spawn_agent tool's visible parameter schema (via tool_search or the tool list) to determine which form is active.

Typed mapping (agent_type-capable schema only):

• Task(subagent_type="X", prompt="Y") → spawn_agent(agent_type="X", message="Y")
• Agent(subagent_type="X", prompt="Y") → spawn_agent(agent_type="X", message="Y")
• Task(model="...") → omit. spawn_agent has no inline model parameter; GSD embeds the resolved per-agent model directly into each agent's .toml at install time so model_overrides from .planning/config.json and ~/.gsd/defaults.json are honored automatically by Codex's agent router.
• Resolved reasoning_effort="low|medium|high|xhigh" (xhigh is a GSD/Codex tier, not a generic runtime enum) → pass reasoning_effort to spawn_agent when the runtime/tool supports it. Omit missing, empty, inherited, or unsupported values; do not invent one-off effort literals in workflow prose.
• fork_context: false by default — GSD agents load their own context via <files_to_read> blocks
• Task(isolation="worktree") / Agent(isolation="worktree") → no direct Codex mapping. Codex spawn_agent does not create or bind a git worktree automatically. Workflows that require this isolation must fail closed or use an explicit manual worktree protocol before spawning (#3360).

Generic-agent workaround (multi_agent_v1 schema — NO agent_type field): When only the generic multi_agent_v1 schema is available, typed GSD agent dispatch (gsd-planner, gsd-executor, etc.) is NOT possible. This is a known Codex limitation (openai/codex#15250). This workaround is NOT equivalent to typed gsd-planner/gsd-executor execution — GSD agents carry project-aware prompts, audit logging, and workflow context that a generic subagent lacks. Use the following fallback:

1. Resolve your active Codex config root — the directory that contains your config.toml. This directory is determined in priority order: $CODEX_HOME (if set), the path given by --config-dir (if passed on invocation), a local .codex directory in the current project (if --local was used), or the default global config directory. Read agents/<agent-name>.toml relative to that config root to extract the agent's system instructions.
2. Inject those instructions as a role-preamble into a generic spawn_agent(message=...) call.
3. Label results and logs clearly as "generic-agent workaround" so the orchestrator and user know full typed-agent guarantees are not in effect.
4. Where typed dispatch is mandatory for correctness (e.g. worktree isolation), fail closed and report the schema limitation rather than silently degrading.

Spawn restriction:

• Codex restricts spawn_agent to cases where the user has explicitly requested sub-agents. When automatic spawning is not permitted, do the work inline in the current agent rather than attempting to force a spawn.
• In some Codex sessions, multi-agent tooling can be deferred. If spawn_agent is not currently visible, discover tools first via tool_search before defaulting to inline execution.

Parallel fan-out:

• Spawn multiple agents → collect agent IDs → wait(ids) for all to complete

Result parsing:

• Look for structured markers in agent output: CHECKPOINT, PLAN COMPLETE, SUMMARY, etc.
• close_agent(id) after collecting results from each agent </codex_skill_adapter>

Conduct a retroactive evaluation coverage audit of a completed AI phase. Checks whether the evaluation strategy from AI-SPEC.md was implemented. Produces EVAL-REVIEW.md with score, verdict, gaps, and remediation plan.

<execution_context> @/home/user/projects/temp/ai-apps/.personal-projects/DSPyTeach/.codex/gsd-core/workflows/eval-review.md @/home/user/projects/temp/ai-apps/.personal-projects/DSPyTeach/.codex/gsd-core/references/ai-evals.md </execution_context>

Phase: {{GSD_ARGS}} — optional, defaults to last completed phase. Execute end-to-end. Preserve all workflow gates.