| name | packet-workflow-skill-builder |
| description | Thin entrypoint for the foundry packet-workflow builder. Use when Codex should invoke the authoritative root builder, contracts, templates, and defaults in this repository. |
Packet Workflow Skill Builder
Thin entry wiring for the foundry's packet-workflow builder.
This subtree is intentionally minimal:
- keep only
SKILL.md and agents/openai.yaml here
- keep authoritative builder logic and tests in
../../../builders/packet-workflow/
- keep authoritative contracts, templates, and default semantics in
../../../core/
- do not reintroduce local copies of contracts, templates, scripts, or tests under this skill subtree
Use this skill only for repo packet workflows that follow collect -> optional lint -> build packets -> optional validate -> optional apply.
Workflow
- Ground the target workflow before scaffolding.
- Inspect the target repo artifacts and existing workflow entrypoints.
- Decide whether the new skill is
audit-only, audit-and-apply, or plan-validate-apply.
- Read
../../../core/contracts/packet-workflow/pattern-catalog.md only when you need help mapping the workflow to the packet pattern.
- Draft a builder spec before generating files.
- Use
../../../builders/packet-workflow/builder-contract.md to lock the builder-spec.json fields.
- Keep the workflow narrow: choose only the packets, scripts, and stop conditions the target workflow actually needs.
- Add
repo_profile only for repo-specific path bindings, review-doc lists, and lint toggles.
- Keep repo profiles declarative and data-only: paths, globs, doc lists, booleans, defaults, and notes only.
- Choose
orchestrator_profile=standard by default and use packet-heavy-orchestrator only when the workflow needs the packet-heavy common path from ../../../core/contracts/packet-workflow/common-path-contract.md.
- Generate the scaffold deterministically.
- Run
python ../../../builders/packet-workflow/scripts/init_packet_skill.py --spec <builder-spec.json> --output-dir <foundry-root>.
- The generator consumes templates from
../../../core/templates/packet-workflow/ and defaults from ../../../core/defaults/packet-workflow/.
- The generator writes the authoritative retained kernel to
../../../builders/packet-workflow/retained-skills/<skill-name>/.
- The generator writes the thin discovery wrapper to
../../../.agents/skills/<skill-name>/.
- Generated skills should treat
../../../.codex/tmp/ as the only repo-local home for temporary, helper, scratch, and ad hoc operator-input files that are not meant for source control.
- Generated skills should place runtime artifacts under the fixed gitignored repo-local root
../../../.codex/tmp/packet-workflow/<skill-name>/<run-id>/.
- Generated skills should default evaluation logging to
<repo-root>/.codex/tmp/evaluation_logs/<skill-name>/<run-id>.json and keep .codex/tmp/ gitignored.
- Validate the generated skill immediately.
- Run
python <codex-home>/skills/.system/skill-creator/scripts/quick_validate.py ../../../.agents/skills/<skill-name>.
- Run
python -m py_compile ../../../builders/packet-workflow/retained-skills/<skill-name>/scripts/*.py for every generated retained script.
Guardrails
- Keep mutation, final synthesis, and broad code changes local in generated skills.
- Keep final adjudication local even when mini workers propose candidate classifications.
- Keep repo profiles data-only end to end; executable behavior stays in scripts and core contracts.
- Treat
worker_selection_guidance as explanatory metadata only; packet_worker_map is the routing authority when present.
- Prefer foundry baseline behavior first, then optional foundry overlay, then project-local profile overrides outside this repo.
- Do not place authoritative builder specs, profiles, references, scripts, tests, or migration worksheets back under
.agents/skills/.
Output
- Tell the user which archetype and packet set you chose.
- Tell the user which generated scripts are placeholders versus ready-to-fill skeletons.
- Tell the user which defaults came from foundry core versus the repo profile scaffold.