Adds an AI-generated, factual German description ("Was macht dieser Skill?")
to scans and shows it in the report.
Changes:
- DB: new nullable `description` column on scansTable (lib/db schema; pushed via drizzle-kit).
- AI: new `generateSkillDescription()` in aiAnalysis.ts — reuses provider selection,
token redaction, system prompt and JSON extraction; expects {"description": "..."},
returns null and never throws on failure.
- Engine: scanEngine now generates the description independently of the AI findings
rules — only a provider+token are required, so it works even when AI findings rules
are disabled. Description failures do not break the scan. EngineResult gains
aiDescription. (Provider/token error precedence unchanged for findings.)
- Prompt: new admin-editable "description" prompt (Beschreibungs-Anweisung) seeded via
onConflictDoNothing, consistent with system/analysis prompts.
- Persist/serialize: description written on scan insert and returned in
serializeScan (list + detail responses).
- API spec: added nullable `description` to the Scan schema in openapi.yaml; regenerated
zod + react-query clients via orval codegen.
- Report UI: new "Was macht dieser Skill?" card in the report header (hidden when empty)
and a matching section in the PDF/print export.
Notes / deviations:
- Old scans are not backfilled (per task scope); their description stays null and the
section is hidden.
- Description is requested as JSON ({"description": ...}) to stay compatible with the
existing "JSON only" system prompt.
- Verified: full typecheck passes, both workflows run, new prompt seeded, scans API
returns description.
Replit-Task-Id: 40c4457b-54d1-4283-a336-478620c3afa8
Verified the AI analysis end-to-end with a real provider and fixed two gaps
found during the live run.
Findings & fixes:
- gpt-5 series (Replit AI Integrations modelfarm default) rejected the hardcoded
`temperature: 0.1` with HTTP 400, silently disabling AI analysis. Removed the
temperature param from the OpenAI-compatible request for broad model
compatibility (aiAnalysis.ts).
- Per-rule AI config (enable/disable/severity) was only a global on/off gate and
AI findings weren't mapped to the AI rule IDs, so individual rule severity was
ignored. runAiAnalysis now receives the enabled AI rules, instructs the model
to classify each finding into one of those ruleIds, drops findings for
disabled rules, and overrides severity/axis with the configured values
(aiAnalysis.ts + scanEngine.ts).
End-to-end verification (Replit OpenAI integration, gpt-5-mini provider):
- "KI-Analyse aktivieren" produces AI findings mapped to AI-PROMPT-INJECTION,
AI-MALICIOUS-INTENT, AI-DATA-PRIVACY.
- Disabling AI-MALICIOUS-INTENT removed its finding; setting AI-PROMPT-INJECTION
to critical was reflected in the result.
- Wrong baseUrl and invalid token (real OpenAI endpoint) produce understandable
aiError messages with no token leak.
Side effects / notes:
- Set up the Replit OpenAI AI Integration (env vars) and created one enabled
provider row ("Replit OpenAI") so AI analysis works out of the box. Each
AI-enabled scan bills the user's Replit credits.
- Test scans created during verification were deleted.
- artifacts/api-server typecheck passes.
Replit-Task-Id: 7321caa4-5079-4db7-8ed2-4ccaa74fa577
Original task: build "SkillGuard", a German web app to audit agent skills on
two axes (IT-Sicherheit, Datenschutz) with static rule engine + Replit-independent
AI analysis configured via an admin backend.
This session:
- Fixed frontend TS errors: lucide-react name collisions (Badge from ui, Activity
from lucide), widened apiType to AiProviderApiType, added queryKey to useGetScan.
- Verified all pages render in German (Dashboard, Prüfen, Bericht, Verlauf, Admin)
and the full scan flow works end-to-end (malicious sample -> verdict block).
Code-review-driven hardening:
- POST /api/scans now returns the full ScanDetail (files + findings) to match the
OpenAPI contract, instead of only the summary.
- AI provider error bodies are redacted (token, Bearer, sk- patterns) before being
returned/persisted, and provider fetches now have a 60s timeout.
- ZIP parsing now enforces limits (max files, total + per-file size) to mitigate
zip-bomb DoS.
Updated replit.md (project overview, decisions, gotchas) and added a memory note
on lucide-react icon name collisions.
