Skills
Repeatable procedures: how to review a migration, implement OAuth, or run a release check.
Skills and Knowledge
Give agents the procedure and the judgment behind it.
A capable agent can produce a plausible change. M6P lets it actively retrieve the approved, company-specific guidance for that change.
They work together
Knowledge
Company-specific decisions, constraints, policies, and rationale: what must be true for that procedure to be safe here.
Where the problem hits hard
When a plausible answer is not enough
Production database changes
An agent can write a migration. It also needs the team’s rollback window, backfill constraints, and service-owner expectations.
Security-sensitive integrations
OAuth, payments, PII, and webhooks carry rules about approved libraries, logging, review, and escalation.
Large monorepos
A change may cross hidden service boundaries, ownership rules, compatibility commitments, and test conventions.
Repeated senior-review feedback
Turn the issues reviewers repeatedly catch into guidance agents can use before human review begins.
Incident fixes under pressure
Keep mitigation, rollback, and communication practices available when speed makes improvisation tempting.
Policy changes in daily work
Make a new security, dependency, or release policy useful at the small decisions where it actually matters.
Vibe coding with teammates
People and agents can move quickly across a shared codebase without every contributor having to rediscover the team’s conventions, boundaries, and hard-won decisions.
Scale senior AI practice to new hires
Senior AI users can turn useful prompts, review habits, decision checks, and team-specific guardrails into Skills and Knowledge that help new teammates contribute with confidence sooner.
Keep the important parts for agents
M6P extracts submitted content in memory and sends the extracted text to the Amazon Bedrock model configured for M6P’s AWS account. M6P does not retain the uploaded file or submitted raw text. It retains the source reference, bounded evidence excerpts, compiler provenance, and reviewable change set. After human review, approved Knowledge can be retrieved alongside approved Skills. See how Knowledge source data is handled.