feat(cli): add openspec agent command group — cut tokens + turns for AI skills

.changeset/agent-helpers.md

@@ -0,0 +1,37 @@
+---
+"@fission-ai/openspec": minor
+---
+
+### Added
+
+- **`openspec resolve-change [name] [--auto] [--json]`** — agent helper that
+  resolves which active change to operate on. Lists every active change,
+  validates a supplied name, or auto-selects when exactly one exists. Distinct
+  exit codes (`0` ok, `1` none, `2` not found, `3` ambiguous) let skills branch
+  without parsing stderr.
+- **`openspec next-artifact --change <name>`** — agent helper that returns the
+  next ready artifact bundled with its full `instructions` payload. Replaces
+  the two-call `status --json` + `instructions <artifact> --json` pattern with
+  one call. JSON by default; pass `--no-json` for a human summary. Emits
+  `{ "done": true }` once every artifact is complete.
+- **`openspec mark-task-done <task-id> --change <name>`** — agent helper that
+  flips a single checkbox in the change's tracking file (typically `tasks.md`)
+  from `- [ ]` to `- [x]`. Idempotent on already-checked lines, preserves
+  CRLF/LF endings, and uses anchored matching so `1.1` does not match `1.10`.
+
+### Changed
+
+- `openspec instructions apply --change <n> --json` payload is now richer:
+  - Each task in the `tasks` array carries an optional `numericId` field
+    (`"1.1"`, `"2.3.4"`) extracted from the start of the task description when
+    the tracking file uses numbered tasks. Existing positional `id` is
+    unchanged.
+  - The top-level payload includes a new `nextPendingId` field: the first
+    unchecked task that has a `numericId`, or `null`. Pair with
+    `openspec mark-task-done` to drive an apply loop without re-parsing
+    `tasks.md`.
+- Skill workflow templates for `openspec-propose` and `openspec-apply-change`
+  now use the three new commands instead of inlining multi-step CLI
+  orchestration. Run `openspec update` against an existing OpenSpec project to
+  pick up the slimmer skill bodies. Net effect: fewer tokens per propose/apply
+  cycle and fewer round-trips between the agent and the CLI.

docs/cli.md

@@ -46,6 +46,9 @@ These commands support `--json` output for programmatic use by AI agents and scr
 | `openspec validate` | Check for issues | `--all --json` for bulk validation |
 | `openspec status` | See artifact progress | `--json` for structured status |
 | `openspec instructions` | Get next steps | `--json` for agent instructions |
+| `openspec resolve-change` | Locate active change | `--auto`/`--json` to drive skills without prompts |
+| `openspec next-artifact` | One-call propose loop | JSON by default; bundles status + instructions |
+| `openspec mark-task-done` | Tick off an applied task | Anchored, idempotent flipping by `numericId` |
 | `openspec templates` | Find template paths | `--json` for path resolution |
 | `openspec schemas` | List available schemas | `--json` for schema discovery |
 | `openspec workspace setup --no-interactive` | Create a workspace with explicit inputs | `--json` for structured setup output |
@@ -858,6 +861,170 @@ openspec instructions design --change add-dark-mode --json
 - Content from dependency artifacts
 - Per-artifact rules from config
 
+When invoked as `instructions apply --change <name> --json`, each task in the
+returned `tasks` array also carries a `numericId` field whenever the tracking
+file uses numbered tasks (e.g. `- [ ] 1.1 Wire foo` → `numericId: "1.1"`). The
+top-level payload also includes `nextPendingId`: the first unchecked task with a
+`numericId`, or `null` when no such task exists. Use this with
+`openspec mark-task-done` to drive an apply loop without re-parsing tasks.md.
+
+---
+
+### `openspec resolve-change`
+
+Agent helper: resolve which active change to operate on. Lists every active
+change, validates a supplied name, or auto-selects when exactly one change
+exists. Useful at the top of every workflow skill where the change identity is
+the first question to answer.
+
+```
+openspec resolve-change [name] [options]
+```
+
+**Arguments:**
+
+| Argument | Required | Description |
+|----------|----------|-------------|
+| `name` | No | Validate that this change is active; echo it on success |
+
+**Options:**
+
+| Option | Description |
+|--------|-------------|
+| `--auto` | Succeed only when exactly one active change exists |
+| `--json` | Output as JSON |
+
+**Examples:**
+
+```bash
+# List every active change as JSON (default when no name and no --auto)
+openspec resolve-change
+
+# Validate a specific name (exit 2 on miss)
+openspec resolve-change add-dark-mode
+
+# Auto-pick the only active change
+openspec resolve-change --auto
+```
+
+**Exit codes:**
+
+| Code | Meaning |
+|------|---------|
+| 0 | Success (name echoed, or JSON list emitted) |
+| 1 | `--auto` and no active changes exist |
+| 2 | Named change not active (or invalid name) |
+| 3 | `--auto` and multiple active changes exist |
+
+---
+
+### `openspec next-artifact`
+
+Agent helper: return the next ready artifact for a change, bundled with the
+full instructions payload for that artifact. Replaces the
+`status --json` + `instructions <artifact> --json` two-call dance with a single
+call. JSON is the default since agents are the primary consumers; pass
+`--no-json` for a human summary.
+
+```
+openspec next-artifact --change <name> [options]
+```
+
+**Options:**
+
+| Option | Description |
+|--------|-------------|
+| `--change <id>` | Change name (required) |
+| `--schema <name>` | Schema override (auto-detected from change config) |
+| `--no-json` | Print a human-readable summary instead of JSON |
+
+**Examples:**
+
+```bash
+# Get the next ready artifact and its instructions (JSON, default)
+openspec next-artifact --change add-dark-mode
+
+# Compact human summary
+openspec next-artifact --change add-dark-mode --no-json
+```
+
+**Output (JSON):**
+
+```json
+// When work remains:
+{
+  "done": false,
+  "artifactId": "design",
+  "resolvedOutputPath": "/abs/path/openspec/changes/add-dark-mode/design.md",
+  "template": "## Overview\n...",
+  "instruction": "Create the design document...",
+  "context": "...",          // optional
+  "rules": ["..."],          // optional
+  "dependencies": [          // completed artifacts to read first
+    { "id": "proposal", "path": "proposal.md", "done": true, "description": "..." }
+  ],
+  "unlocks": ["specs"]
+}
+
+// When every artifact is complete:
+{ "done": true }
+```
+
+**Exit codes:**
+
+| Code | Meaning |
+|------|---------|
+| 0 | Success (next artifact emitted, or `{done: true}`) |
+| 1 | Missing `--change` or change not found |
+| 3 | No runnable artifact — every pending artifact is dependency-blocked |
+
+---
+
+### `openspec mark-task-done`
+
+Agent helper: flip a single checkbox in the change's tracking file (usually
+`tasks.md`) from `- [ ]` to `- [x]`. The target line is identified by the
+leading numeric task id (`1`, `1.1`, `2.3.4`) captured by
+`instructions apply --json`. Idempotent on already-checked lines; preserves
+CRLF vs LF line endings; anchored matching ensures `1.1` does not match
+`1.10`.
+
+```
+openspec mark-task-done <task-id> --change <name> [options]
+```
+
+**Arguments:**
+
+| Argument | Required | Description |
+|----------|----------|-------------|
+| `task-id` | Yes | The hierarchical task id (matches `numericId` from `instructions apply`) |
+
+**Options:**
+
+| Option | Description |
+|--------|-------------|
+| `--change <id>` | Change name (required) |
+| `--schema <name>` | Schema override (auto-detected from change config) |
+| `--json` | Output as JSON |
+
+**Examples:**
+
+```bash
+# Mark task 1.1 done
+openspec mark-task-done --change add-dark-mode 1.1
+
+# Inside an apply loop, driven by nextPendingId from `instructions apply --json`
+TASK=$(openspec instructions apply --change add-dark-mode --json | jq -r .nextPendingId)
+openspec mark-task-done --change add-dark-mode "$TASK"
+```
+
+**Exit codes:**
+
+| Code | Meaning |
+|------|---------|
+| 0 | Success (flipped, or already-done — both are no-op safe) |
+| 2 | Bad input — missing change, schema lacks `apply.tracks`, tracking file missing, or no matching unchecked task line |
+
 ---
 
 ### `openspec templates`

src/cli/index.ts

@@ -32,13 +32,19 @@ import {
   schemasCommand,
   newChangeCommand,
   setChangeCommand,
+  resolveChangeCommand,
+  nextArtifactCommand,
+  markTaskDoneCommand,
   DEFAULT_SCHEMA,
   type StatusOptions,
   type InstructionsOptions,
   type TemplatesOptions,
   type SchemasOptions,
   type NewChangeOptions,
   type SetChangeOptions,
+  type ResolveChangeOptions,
+  type NextArtifactOptions,
+  type MarkTaskDoneOptions,
 } from '../commands/workflow/index.js';
 import { maybeShowTelemetryNotice, trackCommand, shutdown } from '../telemetry/index.js';
 
@@ -532,6 +538,60 @@ newCmd
     }
   });
 
+// Resolve-change command (agent helper: locate active change by name or
+// auto-select when unambiguous)
+program
+  .command('resolve-change [name]')
+  .description('Resolve an active change by name, list active changes, or auto-select the only one')
+  .option('--auto', 'Succeed only when exactly one active change exists')
+  .option('--json', 'Output as JSON')
+  .action(async (name: string | undefined, options: ResolveChangeOptions) => {
+    try {
+      await resolveChangeCommand(name, options);
+    } catch (error) {
+      console.log();
+      ora().fail(`Error: ${(error as Error).message}`);
+      process.exit(1);
+    }
+  });
+
+// Next-artifact command (agent helper: bundle "what is the next ready
+// artifact?" with its full instructions payload). JSON is the default since
+// agents are the primary consumers; pass --no-json for the human summary.
+program
+  .command('next-artifact')
+  .description('Return the next ready artifact for a change, bundled with its instructions (JSON by default)')
+  .option('--change <id>', 'Change name')
+  .option('--schema <name>', 'Schema override (auto-detected from config.yaml)')
+  .option('--no-json', 'Print a human-readable summary instead of JSON')
+  .action(async (options: NextArtifactOptions) => {
+    try {
+      await nextArtifactCommand(options);
+    } catch (error) {
+      console.log();
+      ora().fail(`Error: ${(error as Error).message}`);
+      process.exit(1);
+    }
+  });
+
+// Mark-task-done command (agent helper: flip a tracked tasks.md checkbox by
+// numeric task id)
+program
+  .command('mark-task-done <task-id>')
+  .description('Mark a task complete in the change\'s tracking file (idempotent)')
+  .option('--change <id>', 'Change name')
+  .option('--schema <name>', 'Schema override (auto-detected from config.yaml)')
+  .option('--json', 'Output as JSON')
+  .action(async (taskId: string, options: MarkTaskDoneOptions) => {
+    try {
+      await markTaskDoneCommand(taskId, options);
+    } catch (error) {
+      console.log();
+      ora().fail(`Error: ${(error as Error).message}`);
+      process.exit(1);
+    }
+  });
+
 // Set command group
 const setCmd = program.command('set').description('Set checked-in OpenSpec metadata');
 

src/commands/workflow/index.ts

@@ -22,4 +22,13 @@ export type { NewChangeOptions } from './new-change.js';
 export { setChangeCommand } from './set-change.js';
 export type { SetChangeOptions } from './set-change.js';
 
+export { resolveChangeCommand } from './resolve-change.js';
+export type { ResolveChangeOptions } from './resolve-change.js';
+
+export { nextArtifactCommand } from './next-artifact.js';
+export type { NextArtifactOptions } from './next-artifact.js';
+
+export { markTaskDoneCommand } from './mark-task-done.js';
+export type { MarkTaskDoneOptions } from './mark-task-done.js';
+
 export { DEFAULT_SCHEMA } from './shared.js';

src/commands/workflow/instructions.ts

@@ -229,30 +229,58 @@ export function printInstructionsText(instructions: ArtifactInstructions, isBloc
 
 /**
  * Parses tasks.md content and extracts task items with their completion status.
+ *
+ * When a line begins with a `N` or `N.N(.N)*` token (e.g. `1`, `2.3`, `1.10.4`),
+ * that token is captured as `numericId` and the leading whitespace/punctuation
+ * after it is stripped from `description`. Lines without such a token still
+ * parse, and only get a positional `id`. The positional `id` is always set so
+ * existing callers keep working.
  */
 function parseTasksFile(content: string): TaskItem[] {
   const tasks: TaskItem[] = [];
   const lines = content.split('\n');
   let taskIndex = 0;
 
+  // Capture leading task numbers like `1`, `1.1`, `1.10.4`. Followed by a
+  // separator (space, dot, colon, paren, dash) to anchor against descriptions
+  // that just happen to start with a digit.
+  const numericIdRe = /^(\d+(?:\.\d+)*)(?:[.:)\]-]?\s+|\s+)(.*)$/;
+
   for (const line of lines) {
     // Match checkbox patterns: - [ ] or - [x] or - [X]
     const checkboxMatch = line.match(/^[-*]\s*\[([ xX])\]\s*(.+)\s*$/);
     if (checkboxMatch) {
       taskIndex++;
       const done = checkboxMatch[1].toLowerCase() === 'x';
-      const description = checkboxMatch[2].trim();
-      tasks.push({
+      const rawDescription = checkboxMatch[2].trim();
+      const numericMatch = rawDescription.match(numericIdRe);
+      const task: TaskItem = {
         id: `${taskIndex}`,
-        description,
+        description: numericMatch ? numericMatch[2].trim() : rawDescription,
         done,
-      });
+      };
+      if (numericMatch) {
+        task.numericId = numericMatch[1];
+      }
+      tasks.push(task);
     }
   }
 
   return tasks;
 }
 
+/**
+ * Picks the first unchecked task with a `numericId` (document order). Returns
+ * `null` if no task qualifies. Mirrors what agent skills need to drive
+ * `openspec mark-task-done` without re-parsing the list.
+ */
+function pickNextPendingId(tasks: TaskItem[]): string | null {
+  for (const t of tasks) {
+    if (!t.done && t.numericId) return t.numericId;
+  }
+  return null;
+}
+
 /**
  * Generates apply instructions for implementing tasks from a change.
  * Schema-aware: reads apply phase configuration from schema to determine
@@ -356,6 +384,7 @@ export async function generateApplyInstructions(
     state,
     missingArtifacts: missingArtifacts.length > 0 ? missingArtifacts : undefined,
     instruction,
+    nextPendingId: pickNextPendingId(tasks),
   };
 }