You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Part of the Agentic UX spec series. See sibling issues for related buckets.
Motivation
Agents discover capabilities by parsing --help. That's brittle, verbose, and re-tokenized every session. A single canonical machine-readable catalog gives agents a stable surface to enumerate and lets us evolve help text freely.
Spec
Endpoint
dci commands --json (or dci catalog) emits the entire command tree as one JSON document
Motivation
Agents discover capabilities by parsing
--help. That's brittle, verbose, and re-tokenized every session. A single canonical machine-readable catalog gives agents a stable surface to enumerate and lets us evolve help text freely.Spec
Endpoint
dci commands --json(ordci catalog) emits the entire command tree as one JSON documentSchema (sketch)
{ "version": "1", "cli_version": "1.2.3", "commands": [ { "path": ["anomalies", "recent"], "summary": "List anomalies in a recent window", "flags": [ {"name": "--window", "type": "duration", "default": "24h", "description": "..."}, {"name": "--severity", "type": "enum", "choices": ["low","medium","high"]} ], "output_shape": "AnomalyList", "destructive": false, "requires_auth": true, "agent_friendly": true } ], "shapes": { "AnomalyList": { "...JSON schema..." } } }Properties
versiondestructiveandrequires_authper command (feeds Destructive-operation flagging #17)References
Open questions
dci internal catalog?Acceptance criteria
dci commands --jsonreturns the full surfacedestructiveandrequires_authannotations present per command