Skip to content

Curated Skill Method Sources ​

Purpose ​

Record the external engineering-method sources reviewed for Intentloom's first-party curated skills. This is provenance and compatibility evidence, not an installed extension lock or a legal guarantee.

No third-party plugin, script, asset, hook, or runtime is bundled by the initial curated skill slice. The canonical aif-* wording and control flow are authored for Intentloom.

Reviewed sources ​

SourceReviewed revisionLicenseReviewed areas
obra/superpowers44c9b2d6e889982ac18c27d05a19fefe335194e1, 2026-07-28MITbrainstorming, planning, debugging, TDD, verification, review, worktrees, subagent development, bootstrap and telemetry behavior
mattpocock/skills2ab958093e83e0ec752e6c1c5932da465bf23e0c, 2026-07-28MITgrill flows, domain modeling, bug diagnosis, TDD, planning, implementation, code review, setup, issue-tracker and subagent assumptions

Retained methods ​

  • inspect repository context before design or implementation;
  • use focused questions to expose purpose, constraints, and success criteria;
  • compare materially different approaches and state trade-offs;
  • use reproducible feedback loops and falsifiable debugging hypotheses;
  • apply red-green-refactor where a real behavior seam exists;
  • review standards and specification compliance independently;
  • require fresh verification evidence before completion claims;
  • keep reusable procedures small and composable.

Intentloom adaptations ​

  • extended discovery is risk-based rather than mandatory for every edit;
  • the feature interview never writes or commits unless the user requested it;
  • TDD is required when it provides honest behavior evidence, not when it would manufacture a meaningless failing test;
  • existing Intentloom specifications, ADRs, state, and Duty Watch replace an external CONTEXT.md or plugin-owned documentation hierarchy;
  • Neutron capability and delegation contracts replace plugin-controlled subagent authority;
  • all procedures stop at existing human-approval and transaction boundaries;
  • one canonical catalog generates provider-specific derivatives.

Rejected or deferred mechanics ​

  • mandatory skill invocation based on a minimal probability threshold;
  • mandatory brainstorming and committed design documents for trivial changes;
  • automatic session-start hooks or user-level configuration changes;
  • remote visual-companion assets and version-bearing telemetry;
  • npx ...@latest or another unpinned installer as an adoption path;
  • unconditional commits at the end of an implementation skill;
  • automatic plugin updates;
  • fresh mutating subagents for every plan step;
  • periodic architecture rewrites without a task or roadmap trigger;
  • issue-tracker publication without explicit authorization.

License handling ​

Both reviewed repositories use the MIT License. The initial Intentloom slice adapts engineering concepts and does not copy their scripts, assets, plugin runtime, or substantial skill text. If a future change bundles or substantially copies third-party material, it must preserve the applicable copyright and permission notice, record exact source paths and digests, and pass aif-extension-review before approval.

Update policy ​

These revisions are evidence snapshots, not update subscriptions. A future upstream review must compare the new revision with the pinned source, repeat license and capability analysis, and create a new inactive proposal. No upstream change is inherited automatically.