open/llm-security
1
0
Fork 0
Code Issues Pull requests Releases Wiki Activity Actions
llm-security/scripts/v5-orchestrator-prompt.md
Kjell Tore Guttormsen f153f969a0 feat(ultraplan-local): v1.6.0 — /ultraresearch-local deep research command 
Add /ultraresearch-local for structured research combining local codebase
analysis with external knowledge via parallel agent swarms. Produces research
briefs with triangulation, confidence ratings, and source quality assessment.

New command: /ultraresearch-local with modes --quick, --local, --external, --fg.
New agents: research-orchestrator (opus), docs-researcher, community-researcher,
security-researcher, contrarian-researcher, gemini-bridge (all sonnet).
New template: research-brief-template.md.

Integration: --research flag in /ultraplan-local accepts pre-built research
briefs (up to 3), enriches the interview and exploration phases. Planning
orchestrator cross-references brief findings during synthesis.

Design principle: Context Engineering — right information to right agent at
right time. Research briefs are structured artifacts in the pipeline:
ultraresearch → brief → ultraplan --research → plan → ultraexecute.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-08 08:58:35 +02:00

2.2 KiB
Raw Blame History

Du er orkestrator for llm-security v5.0 "Prompt Injection Hardening".

Plan: /Users/ktg/.claude/plans/ethereal-waddling-rainbow.md Repo: /Users/ktg/.claude/plugins/marketplaces/plugin-marketplace/plugins/llm-security

Oppgaven: Kjør alle 8 sesjoner (S1-S8) sekvensielt via claude -p med --dangerously-skip-permissions. Hver sesjon er en separat headless Claude-invokasjon med fresh context.

For HVER sesjon S1 til S8, gjør dette:

  1. Registrer nåværende HEAD og testcount:

    cd /Users/ktg/.claude/plugins/marketplaces/plugin-marketplace/plugins/llm-security
BEFORE=$(git rev-parse HEAD)
TESTS_BEFORE=$(node --test 2>&1 | grep "^ℹ tests" | awk '{print $3}')

  2. Kjør sesjonen (bytt ut Sn med aktuell sesjon):

    claude -p "Working directory: /Users/ktg/.claude/plugins/marketplaces/plugin-marketplace/plugins/llm-security

Read the v5.0 plan at /Users/ktg/.claude/plans/ethereal-waddling-rainbow.md.
Execute session Sn completely.

Steps:
1. Read the plan's Sn section carefully — every detail matters
2. Implement ALL code changes described there
3. Write ALL tests described there
4. Run: node --test — fix failures until all pass
5. Update CLAUDE.md (test count, hook descriptions) in same commit
6. git add <specific files> && git commit -m 'feat(llm-security): Sn - <description>'
7. git push origin main

Rules:
- Implement ONLY Sn
- All existing tests MUST still pass
- If pathguard blocks Write to settings/hooks, use: write to .tmp then mv
- No subtree push (done at S8 release)
- Report: files changed, tests added, test results" \
  --dangerously-skip-permissions \
  --max-turns 100

  3. Verifiser ETTER hver sesjon:

    • node --test — alle tester passerer (0 failures)
    • git log --oneline -1 — commit finnes og inneholder "llm-security"
    • Testcount økte (forventet: S1 ~45, S2 ~45, S3 ~30, S4 ~50, S5 ~20, S6 ~40, S7 ~10, S8 ~15)

  4. Hvis verifisering feiler: STOPP og rapporter hvilken sesjon som feilet og hvorfor. Ikke prøv neste sesjon.

  5. Hvis verifisering OK: fortsett til neste sesjon.

Etter S8: kjør node scanners/attack-simulator.mjs --verbose og rapporter defense score.

Start nå med S1. Rapporter status etter hver sesjon.