Compound Engineering

The README describes multi-host workflows and says /lfg can implement, simplify, review, commit, and, when a remote exists, push, open a PR, and watch CI. It says merging requires authorization, supporting external effects at 2. The docs also describe some install paths as self-contained, but provide no permission matrix, detailed data flows, sensitive-data handling rules, or general rollback path. A backup appears only in a plan code example and does not establish shipped behavior, so the other trust scores remain conservative. The README names two maintainers, but the object metadata says publisher identity is unverified, and no clear security update responsibility path is shown; source attribution scores 1.

The README gives install and refresh steps for major hosts and mentions cases such as no detected tools and an unparseable settings file, supporting dependency availability at 2 and failure messages at 1. The supplied plans have different dates and statuses, with some marked completed or superseded and others active; their relationship to the current README is not fully verifiable, so self-consistency scores 1. Plan claims about tests are not treated as execution or independent test evidence here.

The material explicitly covers many coding-agent hosts, a core workflow, on-demand skills, and varied installation scenarios, so audience and scenario coverage is thorough. It states boundaries for /lfg, including a bounded repair loop, the possibility of finishing with leftovers at the budget limit, and no merge without authorization; this supports capability boundaries and trigger precision at 2, while host-specific permission and compatibility differences are not comprehensively covered. Host-specific install paths and configuration guidance support environment fit at 3.

The README clearly organizes philosophy, workflow, skill catalog, installation, limitations, FAQ, development, and security/privacy entry points. Installation instructions list commands, refresh steps, and configuration details by host, so information architecture and install notes score 3. Naming is generally patterned, but the material mentions historical renames and legacy-install cleanup, so naming stability scores 2. Examples, FAQ, and limitations exist but are not exhaustive. MIT metadata and license documentation are explicit, scoring 3. The changelog points to release history without including it in the supplied material; maintainers are named and contribution paths exist, but maintenance responsibility and update path are not fully specified, so those scores are 2.

The skills connect brainstorming, planning, implementation, simplification, review, and knowledge capture into a reusable workflow, with directly callable examples; output usability and marginal value score 2. The material also indicates substantial planning and review steps and that /lfg can start a long automated workflow. It provides no cost, time, or outcome comparisons, so cost-benefit scores 1.

The README links skill purposes to guides and identifies SKILL.md as the runtime specification; plan material also names implementation files and acceptance conditions, supporting claim traceability at 2. Most supporting material comes from documents within the same repository, and plan claims about completion and tests lack independent corroboration, so cross-source corroboration scores 1. The README generally separates feature descriptions from usage examples, while plans record assumptions, unverified items, and completion status; this supports fact-inference separation at 2, with the limitation that this review can use only the supplied static excerpts.