Skip to the rows

agentsclimarketplace

Parallel validation

Skill krzemienski/deepest-plan-plugin/skills/parallel-validation

Parallel dispatch patterns for independent validation gates or journeys. Launches multiple sub-agents in a single-message Task burst. Optional skill listed in deepest-plan tool_enrichment when journeys are provably independent.From its SKILL.md

Install

npx -y skills add krzemienski/deepest-plan-plugin --skill parallel-validation

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.0 KB, as published. Nobody here has run it

Parallel Validation

Scope

Speeds up multi-journey validation by dispatching independent journeys in parallel. Only applies when journeys share NO state, NO file mutations, and NO ordering dependencies.

Independence Check (REQUIRED before parallelizing)

All four must be YES:

Check	Question
File independence	Do journeys share files or dirs?
State independence	Do they modify shared state (DB rows, env, auth session)?
Order independence	Does any journey depend on another's output?
Resource independence	Do they contend for a single shared resource (booted simulator, single test user)?

If ANY is NO → serial execution. Do NOT parallelize.

Dispatch Pattern

Single message with N Task tool calls, each targeting one journey:

Message contains N Task tool calls:
  - Journey J1: prompt + evidence paths e2e-evidence/j1-*
  - Journey J2: prompt + evidence paths e2e-evidence/j2-*
  - Journey J3: prompt + evidence paths e2e-evidence/j3-*

All launch simultaneously. Collect results after all complete.

Context Isolation

Each sub-agent receives:

The journey's own steps + PASS criteria (NOT the other journeys')
The shared preflight verdict
The shared validation-plan.md (for cross-reference)
Its own evidence subpath (e2e-evidence/j{N}-*)

Do NOT pass other journeys' evidence or context — causes context bloat and cross-pollination.

Verdict Aggregation

After all sub-agents return:

Read each journey's verdict from its evidence files
Aggregate: OVERALL PASS only if ALL journeys PASS
On any FAIL: invoke error-recovery for only the failing journey(s) — passing journeys stay PASSED

When NOT to Parallelize

Shared booted simulator (only one iOS screenshot at a time)
Shared auth session (logins conflict)
Shared DB row being mutated
Resource-constrained (tight memory, CPU saturated)
Small N (< 3 journeys): orchestration overhead exceeds parallel gain

Integration With Phase 7 Synthesis

Phase 7 (multi-plan tournament) uses this dispatch pattern for:

N parallel plan generators
M parallel judges All at once in single-message bursts.

Rules

Always run the 4-check independence audit before dispatching
Always isolate context per sub-agent (minimal prompts)
Always aggregate verdicts — a single FAIL poisons the whole batch
Always retry only the failing sub-agent(s), not the whole batch
When in doubt, run serial — the orchestration risk isn't worth the speedup

Security Policy

Parallel dispatch shares no state between sub-agents. Each agent's evidence is isolated. Never pass one agent's token / session to another.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.