feat(cli): agent-first WeKnora CLI
The `weknora` command-line client for the WeKnora RAG server. Full command surface: kb / doc / chunk / session / message / agent / model / search / chat / auth / profile / link / api / mcp / skills / config / doctor / schema / exit-codes. Agent-first wire contract: JSON envelope by default, typed error codes mapped to stable exit codes, error.retry_argv / retryable / partial-failure status, global --dry-run preview, the exit-10 confirmation protocol for writes, and control/ANSI/Bidi input hygiene. Stateless env-credential auth (WEKNORA_API_KEY / WEKNORA_TOKEN + WEKNORA_HOST) for headless/CI/agent use. A curated read-only MCP tool surface (`mcp serve`) and embedded Agent Skills (`skills install`). Extensive unit + contract-golden tests and CI.
This commit is contained in:
@@ -1,30 +1,51 @@
|
||||
// Package-level note:
|
||||
//
|
||||
// SetRisk attaches risk metadata to a destructive cobra command via cobra
|
||||
// SetRisk / SetWriteRisk attach risk metadata to a cobra command via cobra
|
||||
// annotations. The SetAgentHelp wrapper in agenthelp.go reads these
|
||||
// annotations and prepends a "Risk: <action> (<level>)" line at the top
|
||||
// of human help output.
|
||||
// of human help output. Destructive ops use SetRisk ("destructive");
|
||||
// reversible update commands use SetWriteRisk ("write"). "read" is reserved.
|
||||
//
|
||||
// envelope.error.risk.action is emitted separately by the
|
||||
// ConfirmDestructive callsite argument and does NOT read these annotations.
|
||||
// ConfirmDestructive / ConfirmWrite callsite argument and does NOT read
|
||||
// these annotations.
|
||||
package cmdutil
|
||||
|
||||
import "github.com/spf13/cobra"
|
||||
|
||||
// RiskDestructive is the only level value currently emitted. The annotation
|
||||
// key reserves "read" / "write" for future use.
|
||||
const RiskDestructive = "destructive"
|
||||
// Risk levels emitted in the annotation / envelope:
|
||||
// - RiskDestructive: irreversible ops (delete, kb init clobber).
|
||||
// - RiskWrite: reversible metadata edits (kb / agent / doc update).
|
||||
//
|
||||
// "read" remains reserved (read-only commands carry no risk annotation).
|
||||
const (
|
||||
RiskDestructive = "destructive"
|
||||
RiskWrite = "write"
|
||||
)
|
||||
|
||||
// SetRisk writes risk metadata to a cobra command's annotations.
|
||||
// SetRisk writes destructive risk metadata to a cobra command's annotations.
|
||||
// Idempotent: re-calling with the same action overwrites cleanly.
|
||||
//
|
||||
// nil-map guard: cobra.Command.Annotations is `map[string]string` and
|
||||
// defaults to nil. Writing to a nil map panics, so we allocate first.
|
||||
func SetRisk(cmd *cobra.Command, action string) {
|
||||
setRisk(cmd, RiskDestructive, action)
|
||||
}
|
||||
|
||||
// SetWriteRisk mirrors SetRisk but tags the command at the "write" level —
|
||||
// used by update commands (kb / agent / doc update), which are reversible
|
||||
// metadata edits rather than irreversible destructive ops. They remain
|
||||
// confirmation-gated; only the level label differs.
|
||||
func SetWriteRisk(cmd *cobra.Command, action string) {
|
||||
setRisk(cmd, RiskWrite, action)
|
||||
}
|
||||
|
||||
// setRisk is the shared writer behind SetRisk / SetWriteRisk.
|
||||
func setRisk(cmd *cobra.Command, level, action string) {
|
||||
if cmd.Annotations == nil {
|
||||
cmd.Annotations = make(map[string]string)
|
||||
}
|
||||
cmd.Annotations["risk.level"] = RiskDestructive
|
||||
cmd.Annotations["risk.level"] = level
|
||||
cmd.Annotations["risk.action"] = action
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user