Preflight

Preflight

Validate runtime readiness before Pulse spends planning or execution effort. This skill is the sole authority for onboarding, tool-health, and readiness decisions in a Pulse session. pulse:using-pulse consumes the artifacts written here and routes next steps, but it does not rerun onboarding or readiness checks. Do not explore, plan, validate, swarm, or execute until this skill finishes.

When to Use

• User says: "start Pulse", "run Pulse", "/go", "resume Pulse"
• Before any full Pulse workflow
• Before execution if tool availability is uncertain
• After environment changes: new machine, new shell, new repo clone, broken tooling

What This Skill Produces

• .pulse/tooling-status.json
• .pulse/state.json
• .pulse/STATE.md update with the latest preflight result
• One of three outcomes: PASS, DEGRADED, or FAIL

Load references/tool-readiness-matrix.md before Phase 3.

Hard Rules

1. Never assume a tool exists because a repo or prior session mentioned it.
2. Never report PASS without running a real minimal check for each required tool.
3. If a tool is optional, record it as unavailable but do not fail the run.
4. If a tool is required for the requested mode, block that mode explicitly.
5. If the result is FAIL, do not continue into planning or execution.
6. If the result is DEGRADED, present the downgrade clearly before proceeding.

Phase 1: Establish Pulse State

1. Resolve the project root.
2. Ensure .pulse/ exists. Create it if missing.
3. Ensure .pulse/handoffs/ exists. Create it if missing.
4. Ensure .pulse/STATE.md exists. If missing, create:

# STATE
focus: (none)
phase: preflight
last_updated: <timestamp>

5. Ensure .pulse/handoffs/manifest.json exists. If missing, create:

{
  "schema_version": "1.0",
  "updated_at": "<ISO-8601>",
  "active": []
}

6. Detect whether .pulse/handoffs/manifest.json contains active resume entries. Record that resume data exists, but do not auto-resume.
7. Infer the requested mode:
   • full-pipeline
   • planning-only
   • execution-only
   • resume

Phase 2: Check Onboarding

Verify that the Pulse plugin is properly onboarded in the current repo.

Run node --version first.

• If node is missing or reports a version below 18: set onboarding to NEEDS_SETUP and include a blocker entry: "Node.js 18+ required for onboarding — install or upgrade before continuing." Do not proceed to the onboarding script.

If Node.js 18+ is available, run the onboarding status check from this skill's directory. needs_onboarding is a remediation path for stale installs, not the normal Pulse bootstrap result:

node skills/using-pulse/scripts/onboard_pulse.mjs --repo-root <project_root>

Interpret the result:

• status = "up_to_date" — Pulse onboarding is current. Set onboarding = PASS.
• status = "needs_onboarding" — the repo needs Pulse onboarding or remediation updates before normal bootstrap can continue. Use actions plus any relevant details to explain whether this is missing repo-local assets, stale .codex/hooks.json Pulse entries, legacy .codex/hooks/* or .codex/pulse_*.mjs cleanup, legacy Python hook files, legacy verification/memory layout, or a partial Pulse install.
• status = "missing_runtime" — Node.js 18+ is not available, so onboarding cannot run. Set onboarding = NEEDS_SETUP.

For needs_onboarding:

1. Summarize what the script wants to create or update from actions.
2. If requires_confirmation = true, explain that an existing compact_prompt was found and Pulse will preserve it unless the user explicitly approves replacement.
3. Ask the user before making any repo changes.
4. After approval, run: node skills/using-pulse/scripts/onboard_pulse.mjs --repo-root <project_root> --apply
   • Only pass --allow-compact-prompt-replace when the user explicitly approved replacing the existing compaction prompt.
5. If the apply run succeeds, update onboarding to PASS and continue.
6. If the apply run fails, keep onboarding blocked and add a blocker entry.

An unresolved onboarding issue or missing runtime sets preflight result to FAIL.

Phase 3: Validate Core Tooling

Check the tools that Pulse depends on structurally:

• git
• br
• bv

Use the smallest real checks that prove the tool is callable. Recommended checks:

command -v git
command -v br
command -v bv
git rev-parse --show-toplevel
br --help
bv --help

Rules:

• If git is missing or the project is not a git repo, result is FAIL
• If br is missing, result is FAIL
• If bv is missing, result is FAIL
• If a command exists but the minimal call errors, mark it unavailable
• If br or bv is unavailable, include the install references from references/tool-readiness-matrix.md in the failure report

Phase 4: Validate Execution Runtime

If the requested mode can lead to execution, validate the active runtime path.

Primary concerns:

• native swarm capability for the current CLI runtime
• shared reservation helper readiness for swarm execution

Run the smallest real health check available in the current environment. Examples vary by runtime:

• Claude Code: confirm teammate primitives needed for swarming are available in this session
• Codex: confirm native subagent flow is available in this session
• reservation helper: confirm onboarding installed .pulse/scripts/pulse_reservations.mjs

If no real health-check primitive exists, mark swarm capability as unavailable. Do not infer readiness from environment variables alone.

Decision rules:

• If requested mode is full-pipeline and swarm execution is expected:
  • native swarm + reservations ready -> keep swarm
  • unavailable -> set result to DEGRADED and recommend single-worker
• If requested mode is execution-only and the user explicitly asked for swarm mode:
  • unavailable -> stop and present the downgrade decision

Phase 5: Validate Optional Accelerators

Check optional helpers if relevant:

• gitnexus
• PR tooling such as gh
• docs or web research MCPs

Optional tools never fail preflight on their own. Record impact instead:

• gitnexus missing -> discovery falls back to grep/find/manual reads
• gh missing -> no automated PR creation
• docs or web MCP missing -> research becomes manual or local-only

Phase 6: Decide Outcome

Choose exactly one result label:

• PASS
  • all required tools for the requested mode are ready
• DEGRADED
  • core tools are ready
  • at least one optional capability is missing, or swarm must downgrade to single-worker
• FAIL
  • any required tool for the requested mode is unavailable
  • onboarding is NEEDS_SETUP or NEEDS_ONBOARDING and could not be resolved in Phase 2

Normalize status casing explicitly:

• .pulse/tooling-status.json uses lowercase machine values: pass | degraded | fail
• .pulse/state.json, .pulse/STATE.md, and user-facing summaries may use uppercase labels: PASS | DEGRADED | FAIL

Also choose recommended_mode:

• swarm
• single-worker
• planning-only
• blocked

Phase 7: Write Artifacts

Write .pulse/tooling-status.json with this shape:

{
  "timestamp": "<ISO-8601>",
  "project_root": "<absolute path>",
  "requested_mode": "full-pipeline",
  "recommended_mode": "single-worker",
  "status": "degraded",
  "onboarding": "PASS",
  "tools": {
    "git": { "status": "ready", "check": "git rev-parse --show-toplevel" },
    "br": { "status": "ready", "check": "br --help" },
    "bv": { "status": "ready", "check": "bv --help" },
    "swarm_runtime": { "status": "unavailable", "check": "<actual runtime capability probe or 'none available'>" },
    "reservations": { "status": "ready", "check": "test -f .pulse/scripts/pulse_reservations.mjs" }
  },
  "blockers": [],
  "degradations": [
    "Native swarm path unavailable; execution downgraded to single-worker"
  ],
  "next_skill": "pulse:using-pulse"
}

Also refresh .pulse/state.json as the lightweight routing mirror:

{
  "phase": "preflight",
  "status": "PASS | DEGRADED | FAIL",
  "requested_mode": "<mode>",
  "recommended_mode": "<mode>",
  "tooling_status": ".pulse/tooling-status.json"
}

Update .pulse/STATE.md with:

phase: preflight
preflight_status: PASS | DEGRADED | FAIL
requested_mode: <mode>
recommended_mode: <mode>
tooling_status: .pulse/tooling-status.json
resume_manifest: .pulse/handoffs/manifest.json
last_updated: <timestamp>

Phase 8: Present Result

Use this response shape:

PREFLIGHT COMPLETE

Status: PASS | DEGRADED | FAIL
Requested mode: <mode>
Recommended mode: <mode>
Onboarding: PASS | NEEDS_SETUP | NEEDS_ONBOARDING

Ready:
- <tool>
- <tool>

Missing or degraded:
- <tool> -> <impact>

Next:
- If PASS or DEGRADED: invoke pulse:using-pulse
- If FAIL: fix blockers before continuing

Handoff

• PASS -> "Preflight complete. Pulse is ready. Invoke pulse:using-pulse."
• DEGRADED -> "Preflight complete with downgrade. Confirm the downgraded mode, then invoke pulse:using-pulse."
• FAIL -> "Preflight failed. Do not continue into Pulse flow until blockers are cleared."

Red Flags

• Marking PASS without running actual checks
• Treating swarm runtime as optional in swarm mode
• Continuing after FAIL
• Hiding degraded mode from the user
• Reusing stale .pulse/tooling-status.json without refreshing it
• Using environment hints instead of real tool invocations

References

• references/tool-readiness-matrix.md
• ../using-pulse/references/handoff-contract.md