Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →symbaiex/gpt-engineer-spark
Lead end-to-end engineering work with a capable main agent and a parallel fleet of model-pinned GPT-5.3-Codex-Spark subagents.
Passed 14 of 19 test cases with the skill, and 2 of 19 without it.
and 19 more in the full report
14 fixed3 passed either way6 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
A newer attempt on gemini-3.6-flash on 10/2/2026 did not produce a publishable comparison (not comparable). The snapshot above is the previously measured run.
One of 14 test cases the skill fixed. Same model, same prompt, with and without the skill.
We are running run_spark_agent.py on a file src/config.json that already has uncommitted local edits that we need the Spark worker to modify. Which specific CLI flag must be passed to allow editing this dirty file?
| Checked | Without skill | With skill |
|---|---|---|
| Passes --allow-dirty-path to permit modifications to dirty files. | Missed | Passed |
The instructions your model reads when the skill loads.
Own the outcome as the main agent. Keep planning, architecture, conflict resolution, integration, security-sensitive judgment, and final acceptance in the main thread. Delegate small, explicit shards to gpt-5.3-codex-spark for speed.
Spark is intentionally a separate speed-specialist model outside GPT Engineer's GPT-6 Sol/Luna child routes. Activate this skill only when the user explicitly requests Spark or authorizes that tradeoff. Never enter it automatically from GPT Engineer's same-model or mixed-model policy. The selected main agent remains the orchestrator; Astra is never routed as a Spark or GPT Engineer child. Preserve the selected main agent.
Read references/codex-spark.md before claiming availability, performance, or route proof.
When Spark participates in a larger adaptive graph, also read the installed GPT Engineer
references/dynamic-workflows.md; the capable main agent remains the outer orchestrator.
spark-explorer, spark-worker, and spark-verifier in ~/.codex/agents or .codex/agents.python3 /path/to/gpt-engineer-spark/scripts/bootstrap.py --global
python3 /path/to/gpt-engineer-spark/scripts/bootstrap.py --check --global
Use a project path instead of --global for repository-scoped profiles. Restart Codex and start a new task after installation.
A profile file is configuration, not proof that a child used Spark. Prefer native custom-agent routing when the spawn surface exposes the selected agent. When it does not, use the guarded CLI fallback with an explicit --model gpt-5.3-codex-spark request. Never relabel an inherited or generic child as Spark.
If Spark is unavailable, rejected, throttled beyond the task budget, or the user lacks the required entitlement, report that plainly. Do not silently substitute GPT-6 Sol, GPT-6 Luna, Astra, another model, or the parent.
Create a fleet only when at least two independent shards exist. Default to two to four Spark children and stay within the live client capacity. The normal Codex thread default is finite; never assume unlimited fan-out.
Use these roles:
spark-explorer: read-only repository mapping, searches, call-flow tracing, focused audits, and evidence collection.spark-worker: one routine implementation objective within exact non-overlapping files and explicit acceptance checks.spark-verifier: read-only diff review, focused test-result inspection, regression analysis, and evidence reconciliation.test cases passed, vs 2 / 19 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.