Skip to content

CHANGELOG.md

Latest commit

 

History

History
337 lines (182 loc) · 12.3 KB

CHANGELOG.md

File metadata and controls

337 lines (182 loc) · 12.3 KB

incur

0.4.2

Patch Changes

  • a6c584d: Fixed stale skills warnings to only appear when synced skill files were still installed on disk, and updated skills list to reflect actual install state instead of stale metadata.

0.4.1

Minor Changes

  • 1e58e47: Breaking: Renamed the global full-envelope flag from --verbose to --full-output, allowing --verbose to be used as a normal command option.

0.3.25

Patch Changes

  • abfa8c7: Fixed Root CLIs created with Cli.create and aliases not registering those aliases as command aliases when mounted via cli.command().

0.3.24

Patch Changes

  • 250e65f: Added command-level aliases option for subcommands (e.g. aliases: ['extensions', 'ext'] on an extension command).
  • 26d7bf8: Fixed root fetch/command fallback bypassing "Did you mean?" suggestions when the input is a typo of a known command.

0.3.23

Patch Changes

  • 572c172: Replaced @readme/openapi-parser with a vendored dereference implementation, removing a heavy dependency tree.

0.3.22

Patch Changes

  • bfc3337: Replaced @modelcontextprotocol/sdk with @modelcontextprotocol/server (v2), reducing dependency count by 74 packages.

0.3.21

Patch Changes

  • d091bf7: Fixed stale skills add CTA commands to use the invoked CLI name when running installed binaries directly, instead of falling back to npx.

0.3.20

Patch Changes

  • ede37be: Fixed help output for boolean options so flags no longer showed <boolean> placeholders or redundant (default: false) text, including aliased flags.
  • 96dfee4: Exported shell completion environment variables in bash and zsh hooks.

0.3.19

Patch Changes

  • 5c76b51: Fixed skills add to list synced skills in alphabetical order.

0.3.18

Patch Changes

  • a65c865: Used command description as fallback in skill frontmatter for root commands and single-command groups.

0.3.17

Patch Changes

  • dad62c9: Fixed skills list not including root command skill.

0.3.16

Patch Changes

  • de70444: Added engines field requiring Node.js >=22.
  • 3462433: Fixed z.bigint(), z.coerce.bigint(), z.date(), and z.coerce.date() schemas failing during skill sync by representing them as { type: "string" } in JSON Schema output.

0.3.15

Patch Changes

  • abd80e7: Fixed missing value errors for flags in Fetch.parseArgv, short secret leaking in redact(), silent jsonl fallthrough in Formatter.format, invalid --format/--token-limit/--token-offset values, lost descriptions when coercing OpenAPI param schemas, and hardcoded process.env in Help.ts for Deno compatibility.
  • 7dd398b: Added skills list subcommand that shows all skills a CLI defines with install status.

0.3.14

Patch Changes

  • 71a787a: Fixed root commands defined on Cli.create() not being included in skill generation (--llms, skills add, .well-known/skills/).

0.3.13

Patch Changes

  • 0e0549f: Added displayName to the root Cli.create(..., { run }) context type.

0.3.12

Patch Changes

  • b8370ac: Added displayName to the run and middleware context. Resolves the actual binary name from process.argv[1] so user-facing messages reflect the alias used to invoke the CLI.

0.3.11

Patch Changes

  • 7833e33: Updated command suggestion ranking to use tiered scoring (prefix → contains → fuzzy) so match type outranks raw edit distance.

0.3.10

Patch Changes

  • d1404b8: Fixed optional properties being typed as required in typegen output.

0.3.9

Patch Changes

  • 8ee1af4: Fixed skill display names to use canonical slug from SKILL.md frontmatter instead of reconstructing from CLI name and subcommand.

0.3.8

Patch Changes

  • 69a48ce: Tweaked "Did you mean" output.

0.3.7

Patch Changes

  • 2f8194b: Added "Did you mean?" suggestions for mistyped commands using Levenshtein distance. Includes builtin commands (mcp, skills, completions) in suggestion candidates. Suggestion CTA preserves original args/flags. Moved skills staleness warning from stderr into the CTA system.

0.3.6

Patch Changes

  • 9b2ab98: Updated help output

0.3.5

Patch Changes

  • 8952a65: Fixed built-in commands (skills, mcp) showing root command errors when invoked without subcommand. Bare skills/mcp and --help now show their own help with available subcommands. Added built-in commands to shell completions. Fixed skill name sanitization for CLI names containing dots.
  • 05d89f3: Fixed resolvePackageRoot failing with ENOENT when running from a Bun compiled binary.
  • 64295d2: Added unified command execution across CLI, HTTP, and MCP transports.

0.3.4

Patch Changes

  • 83aa331: Tweaked help and CTA outputs.

0.3.3

Patch Changes

  • 8adbfbc: Tweaked help output

0.3.2

Patch Changes

  • d2ef65b: Added enum and union-of-literal values in help text instead of generic <value> placeholder.

0.3.1

Patch Changes

  • 1f5a2df: Added support for count options via .meta({ count: true }) on z.number().default(0) schemas. Count flags behave like booleans (no value consumed), but increment on each occurrence, supporting both repeated flags (--verbose --verbose) and stacked aliases (-vvv).

0.3.0

Minor Changes

  • 9add1a0: Breaking: Renamed --llms to --llms-full. Added a new --llms flag that outputs a compact command index (table of command signatures + descriptions) instead of the full manifest. This reduced token usage by ~95% for agents that already know the CLI and just need a quick reminder of available commands.

Patch Changes

  • a2610bc: Added requires_bin and fallback descriptions to generated skill frontmatter.
  • dd7a1af: Fixed --no-global resolving cwd to the CLI's installation directory instead of process.cwd().

0.2.2

Patch Changes

  • 9454412: Added --token-count, --token-limit, and --token-offset global options for token-aware output pagination. Uses LLM tokenization estimation (~96% accuracy via tokenx). In --verbose mode, truncated output includes meta.nextOffset for programmatic pagination.

0.2.1

Patch Changes

  • 6ab9a33: Added --filter-output global option to filter output by key paths with support for dot notation and array slicing.
  • 2dc1b00: Added --schema global option to every command that returns its JSON Schema (args, env, options, output).
  • c60e6b8: Exposed format and formatExplicit on run and middleware context.
  • 0e52ec0: Added cli.fetch to expose CLI as a standard Fetch API handler
  • f5b0133: Added optional exitCode to c.error() and IncurError, allowing CLI authors to control the process exit code. Defaults to 1 when omitted (backward compatible).

0.2.0

Minor Changes

  • 00b0b2d: Added Fetch API integration — mount any HTTP server as a CLI command.
    • Fetch gateway: .command('api', { fetch: app.fetch }) translates argv into HTTP requests using curl-style flags (-X, -d, -H, --key value query params)
    • Streaming: NDJSON responses (application/x-ndjson) are streamed incrementally
    • OpenAPI support: .command('api', { fetch, openapi: spec }) generates typed subcommands with args, options, and descriptions from an OpenAPI 3.x spec
    • Works with any framework exposing a Web Fetch API handler (Hono, Elysia, etc.)

0.1.17

Patch Changes

  • b73feaf: Added aliases option to Cli.create for registering alternative binary names. Shell completions and help output include all aliases.

0.1.16

Patch Changes

  • e3aa038: Added dynamic shell completions for bash, zsh, fish, and nushell. CLIs get a built-in completions <shell> command that outputs a hook script. The hook calls back into the binary at every tab press, so completions stay in sync with commands automatically. Supports subcommands, --options, short aliases, enum values, and space suppression for command groups.
  • 06580f0: Added short-alias stacking (e.g. -abc parsed as -a -b -c). The last flag in a stack can consume a value; all preceding flags must be boolean.

0.1.15

Patch Changes

  • 5122c9b: Fixed help formatter using process.env instead of env source override for "set:" display

0.1.14

Patch Changes

  • 3f7ca73: Added leading # to CTA command descriptions for easier copy-paste.
  • 3f7ca73: Moved environment variables section to bottom of help output.
  • 3f7ca73: Fixed invalid subcommand in a group falling through to root handler instead of returning COMMAND_NOT_FOUND. Added CTA with copyable help command to COMMAND_NOT_FOUND errors.
  • 50282a8: Added redacted current value indicator for environment variables in help output.
  • 79fbabd: Fixed streaming handler ignoring CLI-level and command-level default format. Previously, handleStreaming used only formatExplicit to decide between incremental and buffered mode, causing CLI defaults like { format: 'json' } to be ignored in favor of hardcoded 'toon'.

0.1.13

Patch Changes

  • aa32795: Added version to the command run context (c.version).

0.1.12

Patch Changes

  • a61c474: Added help output in human mode for root command with args when no args provided

0.1.11

Patch Changes

  • 77f5c98: Added deprecated option support via Zod's .meta({ deprecated: true }). Deprecated flags show [deprecated] in help output, **Deprecated.** in skill docs, deprecated: true in JSON Schema, and emit stderr warnings in TTY mode.

0.1.10

Patch Changes

  • e7564a0: Added c.error() to middleware context for structured error short-circuiting. Middleware can now return c.error({ code, message }) instead of throwing, producing a proper error envelope with optional CTAs.

0.1.9

Patch Changes

  • 1a671e9: Added name to run and middleware context (c.name) — returns the CLI name passed to Cli.create().

0.1.8

Patch Changes

  • eec5906: Added c.env to middleware context. CLI-level env schema defined on Cli.create() is now parsed before middleware runs and available as typed c.env in both .use() and per-command middleware: [...] handlers. This enables initializing shared dependencies (API clients, auth tokens) in middleware using validated environment variables instead of reading process.env directly.

0.1.7

Patch Changes

  • 2c60110: - Added middleware support via cli.use().
    • Added typed dependency injection via vars: declare a Zod schema on create() (and optionally set defaults), set values with c.set() in middleware, read them via c.var in handlers.
  • ba07f0b: Added per-command middleware via middleware property on command definitions. Added middleware() helper for creating strictly typed middleware handlers with middleware<typeof cli.vars>(...). Added cli.vars property to expose the vars schema for use with typeof.

0.1.6

Patch Changes

  • 6642c48: Added agent boolean to the run context. true when stdout is not a TTY (piped/agent consumer), false when running in a terminal. Use it to tailor command behavior for agents vs humans.
  • 6642c48: Added outputPolicy option to commands, groups, and root CLIs. Set outputPolicy: 'agent-only' to suppress data output in human/TTY mode while still returning structured data to agents. Defaults to 'all'. Inherited from parent groups — children can override.

0.1.5

Patch Changes

  • b334523: Added automatic cleanup of stale skills when commands are removed or depth changes. Fixed broken symlinks not being removed on Node v24.

0.1.4

Patch Changes

  • 9bb41e3: Fixed --depth=N equals syntax not being parsed in skills add. Fixed depth=0 producing a root SKILL.md without a subdirectory wrapper.

0.1.3

Patch Changes

  • 0e42bc0: Added native skill installation.

0.1.2

Patch Changes

  • dfd804c: Added ability for a root command to have both a run handler and subcommands. Subcommands take precedence — unmatched tokens fall back to the root handler. --help shows both root command usage and the subcommand list.

0.1.1

Patch Changes

  • 370d039: Fixed commands returning undefined being serialized as the literal string "undefined" in output. Void commands now produce no output in human and machine modes. MCP tool calls with undefined results now return valid JSON (null) instead of broken output.

0.1.0

Minor Changes

  • 09e4d76: Initial release.

0.0.2

Patch Changes

  • 9c7f8aa: Updated SKILL.md
  • 3d38f2d: Added usage info at end of description frontmatter in skills.

0.0.1

Patch Changes

  • 1318c14: Initial release