Benchmark external agent frameworks, auto-detect source type, scan for prompt injection, and convert findings into a concrete TDD upgrade backlog for agent-studio evolution.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Benchmark external agent frameworks, auto-detect source type, scan for prompt injection, and convert findings into a concrete TDD upgrade backlog for agent-studio evolution.
Skipping injection scan on external content — always scan before analysis
Ignoring source type detection — auto-detect reduces misrouted analysis
Source Auto-Detection (Inspired by Skill_Seekers SourceDetector)
When the input source is ambiguous, auto-classify before proceeding:
Input Pattern
Source Type
Analysis Strategy
https://github.com/owner/repo
GitHub repo
Three-stream: code + docs + community
owner/repo (no URL)
GitHub shorthand
Clone via git clone --depth=1
https://... (non-GitHub URL)
Documentation site
Web scrape + structure extraction
Local directory path
Local codebase
Direct file analysis
*.pdf, *.docx, *.epub
Document file
Content extraction pipeline
*.json, *.yaml config
Config/manifest
Schema + structure analysis
PyPI/npm package name
Package registry
Fetch metadata + clone source
Decision tree: Check GitHub URL → check file extension → check if local path exists → check if package name → fall back to web URL.
Write detected source info to <run-id>/source-info.json:
{"type":"github|web|local|document|package","parsed":{"url":"...","owner":"...","repo":"..."},"suggestedName":"auto-generated-name","rawInput":"original user input"}
On resume: read progress file → skip completed phases → continue from nextStep.
Benchmark Comparison Report (Inspired by Skill_Seekers BenchmarkRunner)
After Phase 3, generate a structured comparison report at <run-id>/comparison-report.json:
{"name":"agent-studio vs <external-repo>","comparedAt":"<ISO>","dimensions":[{"dimension":"memory_model|search_stack|agent_orchestration|creator_system|observability|security|testing|documentation","ours":{"description":"...","maturity":"none|basic|intermediate|advanced"},"theirs":{"description":"...","maturity":"none|basic|intermediate|advanced"},"verdict":"ahead|parity|behind|different_approach","adoptionCandidate":true}],"summary":{"totalDimensions":8,"ahead":0,"parity":0,"behind":0,"differentApproach":0,"adoptionCandidates":0},"topFindings":["...","..."],"injectionScanPassed":true}
This replaces ad-hoc prose comparison with a machine-readable format that enables tracking improvements over time and across multiple assimilation runs.
Workflow Template Support (Inspired by Skill_Seekers YAML Workflows)
When the external project uses composable workflow definitions (YAML, JSON, or similar), extract the workflow pattern and document it in <run-id>/workflow-patterns.md:
Stage definitions — what stages exist, their types (builtin vs custom), and ordering
History chaining — which stages consume output from previous stages (uses_history: true)
Post-processing — any section reordering, metadata injection, or cleanup steps
Variables — configurable parameters that modify workflow behavior
This analysis feeds into the gap list — if our framework lacks composable stage-based workflows for a given domain, that becomes a gap candidate.
Memory Protocol (MANDATORY)
Before work: cat .claude/context/memory/learnings.md
After work: record assimilated patterns → learnings.md; adoption risks → decisions.md; blockers → issues.md.
