fullsend experiments

Experiments for the fullsend platform — each tests a hypothesis about autonomous agent infrastructure, security, tooling, or workflows.

Experiments

#	Experiment	Status
0001	Agent outage fire drill	Active
0002	Claude-based ADR drift scanner	Concluded
0003	ADR-0046 drift scanner	Concluded
0004	Zero-config autonomous bug fix engine	Concluded
0005	Agent scoped tools triage	Concluded
0006	Code agent evaluation	Concluded
0007	GitHub Actions agent runtime MVP	Concluded
0008	Guardrails evaluation	Concluded
0009	Hermes-inspired security patterns	Concluded
0010	Host-side API server for sandboxed agents	Concluded
0011	Integration Service design doc drift	Concluded
0012	Model Armor vs AI agent triage	Concluded
0013	OpenShell policy bypass	Concluded
0014	OpenShell sandbox evaluation	Concluded
0015	Prompt injection defense-in-depth	Concluded
0016	Promptfoo for agent evaluation in CI	Concluded
0017	Reasoning monitor	Active
0018	Runner hello world	Active
0019	Skills	Active
0020	Target repository skills in triage	Concluded
0021	Tool scoping	Concluded
0022	Claude GitHub App auth	Concluded
0023	Review cache publication policy	Concluded

Conventions

Experiments follow a numbered directory convention. See AGENTS.md for full details.

• Naming: NNNN-short-description/ (zero-padded 4-digit number)
• Frontmatter: YAML with title, status, and optional topics
• Statuses: Active, Concluded, Abandoned, Merged
• Template: 0000-experiment-template
• Linting: hack/lint-experiment-numbers and hack/lint-experiment-frontmatter enforce conventions via pre-commit