Inventory
Known roots, active resource types, target presence, config surfaces, and script-bearing skills.
inspect.targets[]
v0.1 · read-only inspector
Your agents have an environment. Make it legible.
Inspect the skills, guidance, agents, commands, hooks, MCP surfaces, and possible secret-bearing fields distributed across your AI tools— without changing a file.
curl -fsSL https://agentctl.justrepl.com/install.sh | bash
Installs to ~/.local without sudo. Verifies the release checksum.
$ agentctl inspect
agentctl inspect · read-only
4/5 targets detected · 83 resources
TARGETSTATESKILLS
Codexdetected28
Claude Codedetected27
Gemini CLIdetected12
OpenCodedetected14
Cursorabsent0
HIGH 2 potential secret-bearing fields detected
MED 1 same-name skill has divergent definitions
INFO 15 byte-identical skills can share source
No files were changed. Use --json for the full report.
Codex
Claude Code
Gemini CLI
OpenCode
Cursor
Future adapters
Observe before manage
One scan. Four kinds of signal.
agentctl does not begin by asking for authority. It begins by explaining the environment that already exists.
Duplication
Byte-identical skill definitions that can be consolidated into one canonical source.
sha256(SKILL.md)
Divergence
Same-name resources whose definitions differ across targets and need an explicit decision.
name → variants
Risk surface
Potential secret keys, hooks, MCP configuration, and executable files—without printing values or running code.
mode: read-only
Agent Environment as Code
File sync is not semantic control.
A file can arrive at the right path and still be ignored, weakened, or interpreted under a different scope. agentctl keeps the evidence layers separate.
- 01 Desired canonical intent
- 02 Rendered target-native output
- 03 Filesystem observed in v0.1
- 04 Runtime target discovery
- 05 Behavior bounded evidence
v0.1 observes filesystem truth. It never claims that static presence proves runtime discovery or model behavior.
The honest release
Useful now. Explicit about next.
v0.1.1 alpha
Shipping
- Read-only multi-agent inventory
- Duplicate and divergence analysis
- Secret-key and executable-surface signals
- Human and stable JSON output
- Safe canonical repository scaffolding
- Checksum-verified, no-sudo installation
Roadmap
- Target-native semantic adapters
- Capability-loss accounting
- Content-addressed plan and apply
- Atomic rollback and drift records
- Secret-manager references
- Runtime and behavioral conformance
Trust is product behavior
Inspection should not require surrender.
No writes
inspect and doctor never modify agent roots.
No execution
Targets, plugins, skills, scripts, and hooks are never invoked during inspection.
No secret values
Reports contain potential key names and line numbers—not detected credential values.
No telemetry
The CLI has no analytics endpoint, account requirement, or inspection upload.
Open source · Apache 2.0