dreaming — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited dreaming (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.
Turn yesterday's execution into tomorrow's leverage.
This is a read-only synthesis skill. It studies recent work across repos and agent artifacts, asks better research questions, distills what was learned from project and coding execution, and writes a compact morning report. It must not mutate production code, create tickets, push branches, or expose secrets unless the user explicitly asks for a follow-up execution phase.
Use dreaming when the work is already done or partially done and the next useful act is reflection:
Default scope:
.planning/, .gsd/, .omc/, handoff files, and reports;AGENTS.md, CLAUDE.md, README.md, ARCHITECTURE.md, and existing codebase maps;dreaming is the reflection layer around AOSwarm. It does not make code changes itself. When it discovers concrete implementation work, convert that into AOSwarm-ready prompts:
Use AOSwarm for follow-up when a lesson becomes executable code work. Keep the dream report as the decision surface.
Default to the last 24 hours for nightly runs and the last 7 days for ad hoc "catch me up" runs. State the window explicitly in the report.
Read only. Prefer cheap structured commands first:
git status --short
git log --since="24 hours ago" --oneline --decorate
git branch --show-current
find . -maxdepth 3 \( -name AGENTS.md -o -name CLAUDE.md -o -name README.md -o -name ARCHITECTURE.md \)For multi-repo runs, rank repos by recent commits, dirty state, branch activity, and recently modified planning/report artifacts. Do not scan node_modules, .git, virtualenvs, build outputs, or huge generated folders.
For each active repo, answer:
Prefer lessons that improve future throughput across tasks. Good insights look like:
Reject vague observations such as "write better tests" unless they point to a specific contract, missing fixture, or failure mode.
Organize questions by category:
| Category | Question shape |
|---|---|
| Architecture | "What boundary would reduce repeated cross-module edits?" |
| Testing | "Which invariant needs a contract test so agents stop guessing?" |
| Tooling | "What script would collapse repeated setup/debug time?" |
| Product semantics | "Which business rule is implicit in code but absent from specs?" |
| Agent workflow | "Where did context compaction, permissions, or routing slow work?" |
| AOSwarm design | "Which follow-up tasks are independent enough for parallel workers?" |
| External research | "What do current docs, GitHub examples, or X/Twitter builders suggest?" |
When using public research, treat X/Twitter as discovery only. Verify useful claims against primary sources, GitHub repos, docs, code, or reproducible local experiments before turning them into engineering recommendations.
Write both Markdown and HTML when running unattended:
reports/nightly/YYYY-MM-DD/dreaming.md
reports/nightly/YYYY-MM-DD/index.htmlRequired report sections:
HTML must be self-contained, light-mode, and openable by double-clicking. Use the html-reports skill conventions.
If the primary model fails with rate-limit, quota, overload, or authentication errors:
Never loop indefinitely. After three consecutive provider failures, mark that provider unavailable for the run.
.env; if a repo requires environment loading, use its approved loader and never print secrets.A dreaming run is complete when it has:
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.