learn-existing — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited learn-existing (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
The job: load what's needed to act correctly — no more. The core tension is fidelity vs context rot, resolved by one rule:
Primary artifacts are read directly; surveys are delegated. A plan file about to be executed, a PRD driving a design, the key strategy note — read these verbatim into the main context (summarization loses execution-critical detail). Broad terrain — a whole codebase, the vault at large, a long CI log — goes to an Explore subagent that returns a structured summary (relevant files, conventions, reusable pieces, constraints, risks). Never paste raw survey output into the main context.
The caller passes a focus (combinable, e.g. vault+codebase). Invoked bare, infer the focus from the conversation (a PR number present? a PRD named? a plan folder mentioned?) and say which one you picked.
| Focus | Load directly | Delegate to subagent |
|---|---|---|
vault | The 2–4 most relevant notes from the map in references/vault-map.md | Wider vault exploration for the topic |
codebase | The repo's CLAUDE.md + path-scoped .claude/rules/* relevant to the task | Full-repo analysis: structure, conventions, design system, reusable pieces, constraints |
prd:<name> | The PRD + ALL its Initiative files (verbatim) | Related/competing PRDs check |
plans:<folder> | The target plan file(s) + plan/INDEX.md + the YYYY-MM-DD-design.md | Codebase reality-check of the plan's named paths |
pr:<number> | gh pr view metadata + checks + linked plan/initiative | Diff reading when >300 lines; failing CI log extraction (gh run view --log-failed) |
Read references/vault-map.md before any vault or prd: focus — it maps where everything lives and the conventions that govern writes (frontmatter taxonomy, naming, Index.md). Also consult its repo map (§5) and brainstorm-anatomy (§6) sections when resolving codebase or plans: targets — it is the only place all 6 repo paths are recorded.
End with a compact grounding statement so the caller can trust the loaded state:
CONTEXT LOADED
Sources: <what was read directly / what was surveyed>
Know: <3–6 bullets of load-bearing facts>
Unknown: <what remains open and would need asking or deeper digging>Keep it honest — an "Unknown" line that names a real gap is worth more than false confidence. Don't re-load what's already in context from earlier in the session; say "already loaded" instead.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.