goal-preflight

name	goal-preflight
version	0.2.117
description	Prepare path-hardened /goal orchestration bundles from a report, roadmap, diagnosis, or rough goal brief. Use when the user needs prompt infrastructure for goal-main-orchestrator: optionally use CLI-only Lite advisors for source digestion or lint-repair advice, synthesize rolling-scheduled branch groups and worker-sized work items when missing, enforce reproducible manifest paths and telemetry requirements, write job.manifest.json/main.prompt.md/branch prompts/location-bound goal-bootloader.md, run deterministic lint, and present the exact bootloader text for manual /goal launch.

Goal Preflight

Prompt-prep wrapper only. Do not launch /goal, runtime auditors, branch orchestrators, reviewers, workers, or plan-amenders.

Start

Resolve the installed skills root:

GOAL_SKILLS_ROOT="${CODEX_HOME:-$HOME/.codex}/skills"
if [ ! -d "$GOAL_SKILLS_ROOT/goal-preflight" ] && [ -d "$HOME/.agents/skills/goal-preflight" ]; then
  GOAL_SKILLS_ROOT="$HOME/.agents/skills"
fi

Then print the compact deterministic phase table and follow it:

python3 "$GOAL_SKILLS_ROOT/goal-preflight/scripts/runtime_phase_manifest.py" --markdown

For a normal new bundle, run the guided pipeline after writing the brief:

python3 "$GOAL_SKILLS_ROOT/goal-preflight/scripts/prepare_goal_bundle.py" \
  --brief /abs/brief.json \
  --repo-root /abs/repo \
  --out-dir /abs/bundle \
  --output /abs/bundle/preflight.pipeline.json

For a compact state snapshot (no jq), run:

python3 "$GOAL_SKILLS_ROOT/goal-preflight/scripts/render_goal_bootloader.py" --bundle-dir /abs/bundle --readiness

Use --json for machine-readable readiness checks. The guided pipeline writes preflight.brief.lint.json, preflight.lint.json, repair-gate.json, readiness.json, goal-config-selection.json, and compact preflight.pipeline.json; add prepare_goal_bundle.py --verbose only when full embedded selection/readiness payloads are needed.

For compact prompt handoff, share only the --readiness --json payload and next-command guidance instead of full manifests.

Runtime Rules

• Produce a structured brief JSON, then let scripts generate and lint the bundle.
• If the user asks for debug mode or debug telemetry, set telemetry_mode: "debug" in the structured brief; bundle creation expands the full safe debug telemetry policy.
• If the user supplied or requested a goal configuration, prefer prepare_goal_bundle.py; it auto-detects candidate configs, runs check_goal_config.py --for-preflight, writes remediation output when caps/telemetry fields can be sanitized, and records the decision in goal-config-selection.json. Use manual --goal-config /abs/goal.config.json --goal-config-check /abs/goal-config-check.json only when debugging a specific stage.
• If brief shape is unclear, run create_goal_bundle.py --brief-schema-json or --example-brief; do not inspect script source for schema.
• Parallelism is default: prefer independent branches and worker-sized work items; record serial reasons when capacity is intentionally underfilled.
• When preparing a new bundle, preserve intermediate artifacts and use the guided preflight pipeline. If a manual fallback is needed, run brief lint, create_goal_bundle, lint_goal_bundle, script_only_repair_gate, then render_goal_bootloader with --readiness before returning bootloader text.
• Readiness must treat non-git repository roots as blocked for runtime branch/worktree orchestration unless a future explicit no-git runtime mode is implemented; blocked bootloaders are corrective-only and must not render $goal-main-orchestrator launch handoff commands.
• Ask the user only for gaps that would change branch boundaries, DoD, merge policy, or runtime safety.
• Use Lite only as optional context routing for large/vague source material or lint repair.
• Return the exact goal-bootloader.md text after lint passes and readiness launch is allowed; if readiness is blocked, return the blocked readiness fix/recheck command instead.
• Do not read or search skills/*/scripts/*.py during normal preflight, including with rg, grep, cat, sed, or head. Inspect Python source only when a script failed and debugging that script is the assigned task.

Details On Demand

Open detailed references only after a phase script or linter points at an ambiguity:

• references/actionability-rubric.md for vague source material.
• references/bundle-contract.md for bundle lint defects.
• references/parallelization-rules.md for decomposition tradeoffs.
• references/lite-advisor-contract.md before creating Lite packets.