How g26karthik/ghost-data-still-exists works
Ghost Data Still Exists audits deletion requests: it parses evidence, resolves historical identities, reconstructs resource state at the audit cutoff, and produces a policy-based verdict and ordered plan. The main entry point also supports an agent workflow in which a model investigates through read-only tools and submits a plan for an independent gate. The graph preserves these documented stages and major capabilities. Some implementation paths are unsampled, so links not established by excerpts or README statements are omitted or marked unverified.
Main entry points: main.py Audit runner · agent.py Model investigator · identity.py Identity index · planner.py Verdict and plan
Components 17
Audit workflow
- Audit runner
main.py - Request intake
ghost/intake.py
Evidence and state
- Evidence parser
ghost/evidence.py - Screenshot transcription
ghost/vision.py - LLM client
ghost/llm.py - Identity index
ghost/identity.py - Worker correlation
ghost/activity.py - Cutoff state
ghost/state.py
Decision and gate
- Deletion policy
ghost/policy.py - Verdict and plan
ghost/planner.py - Output validator
ghost/validate.py
Agent investigation
- Model investigator
ghost/agent.py - Read-only tools and gate
ghost/agent_tools.py
Other components
- Audit operator
- Request and evidence files
- Validated audit output
- OpenAI API
Connections 21
- Audit operator → Audit runner: starts audit
README.md - Request and evidence files → Audit runner: provides dataset
README.md - Audit runner → Evidence parser: parses evidence
main.py - Audit runner → Identity index: resolves identity
main.py - Audit runner → Worker correlation: assigns worker actions
main.py - Audit runner → Cutoff state: reconstructs cutoff state
README.md - Audit runner → Verdict and plan: builds plan
main.py - Audit runner → Deletion policy: loads rules
main.py - Audit runner → Request intake: runs intake
main.py - Screenshot transcription → LLM client: requests transcription
ghost/vision.py - Evidence parser → Identity index: extracts identifier tokens
ghost/evidence.py - Identity index → Worker correlation: supplies identity links
ghost/activity.py - Verdict and plan → Deletion policy: applies treatment rules
ghost/planner.py - Audit runner → Model investigator: starts agent mode
main.py - Model investigator → Read-only tools and gate: invokes tools
README.md - Read-only tools and gate → Verdict and plan: audits proposed plan
ghost/agent_tools.py - Model investigator → OpenAI API: calls model API
ghost/agent.py - LLM client → OpenAI API: calls Responses API
ghost/llm.py - Audit runner → Output validator: validates output
main.py - Output validator → Validated audit output: checks contract
ghost/validate.py - Audit runner → Validated audit output: writes audit results