Not full text, but:

- config you cli to allow run agents in parallel

- ask to use tmp dir for all files, and git worktree copy

- ask to export all to files, progress, errors, reports, as it goes.

Now setup for each agent run (process with own context):

- planning-with-files, caveman, ask-questions-if-underspecified(if not in CI)

- these skills must be loaded into each subagent.

After, run assembling context: - audit-context-building, trailmark-structural

At this point all only operator runs. No other agents.

Now ask for forks(parallel), each loads context built before:

- dba-review

- supply-chain-risk-auditor (ask to trigger only on lock files changes

- spec-to-code-compliance

- openai-gh-fix-ci (if audit of PR)

- graph-evolution, differential-review

- dimensional-analysis

- mutation-testing, property-based-testing (ask to mutate and proptest around changes).

So not of these skills are load into shared context. They have own to avoid pollution and loss of focus.

So all forks run done. Ask for join.

- fp-check, second-opinion

Ask for final report.

All skills above from github and public (not my, easy to find).

I have private skills like:

- github-context-aggregation (links, comments, descriptions)

- structural-transformation-review (thing topology, category theory, duals)

- layered-context-chunking (2 layers at a time - like db+backend, backend+fe to more focus)

- product-feature-interactions (product line eng combinations)

Good mental model for what you are doing is

Aikido Code Audit