guild-agents 1.5.0 → 2.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (51) hide show
  1. package/README.md +71 -67
  2. package/bin/guild.js +4 -85
  3. package/package.json +1 -1
  4. package/src/commands/doctor.js +11 -33
  5. package/src/commands/init.js +1 -1
  6. package/src/templates/skills/build-feature/SKILL.md +7 -38
  7. package/src/templates/skills/build-feature/evals/evals.json +2 -2
  8. package/src/templates/skills/council/SKILL.md +4 -14
  9. package/src/templates/skills/council/evals/evals.json +3 -13
  10. package/src/templates/skills/create-pr/SKILL.md +2 -5
  11. package/src/templates/skills/guild-specialize/SKILL.md +2 -5
  12. package/src/templates/skills/qa-cycle/SKILL.md +0 -7
  13. package/src/templates/skills/re-specialize/SKILL.md +0 -3
  14. package/src/templates/skills/session-end/SKILL.md +77 -30
  15. package/src/templates/skills/session-start/SKILL.md +51 -20
  16. package/src/utils/eval-runner.js +2 -8
  17. package/src/utils/generators.js +3 -4
  18. package/src/utils/skill-parser.js +83 -0
  19. package/src/utils/trigger-runner.js +1 -1
  20. package/src/commands/logs.js +0 -63
  21. package/src/commands/reset-learnings.js +0 -44
  22. package/src/commands/run.js +0 -105
  23. package/src/commands/stats.js +0 -147
  24. package/src/templates/agents/learnings-extractor.md +0 -49
  25. package/src/templates/skills/dev-flow/SKILL.md +0 -81
  26. package/src/templates/skills/dev-flow/evals/evals.json +0 -36
  27. package/src/templates/skills/dev-flow/evals/triggers.json +0 -16
  28. package/src/templates/skills/new-feature/SKILL.md +0 -119
  29. package/src/templates/skills/new-feature/evals/evals.json +0 -41
  30. package/src/templates/skills/new-feature/evals/triggers.json +0 -16
  31. package/src/templates/skills/review/SKILL.md +0 -97
  32. package/src/templates/skills/review/evals/evals.json +0 -43
  33. package/src/templates/skills/review/evals/triggers.json +0 -16
  34. package/src/templates/skills/status/SKILL.md +0 -100
  35. package/src/templates/skills/status/evals/evals.json +0 -40
  36. package/src/templates/skills/status/evals/triggers.json +0 -16
  37. package/src/templates/skills/verify/SKILL.md +0 -114
  38. package/src/templates/skills/verify/evals/triggers.json +0 -16
  39. package/src/utils/accounting.js +0 -139
  40. package/src/utils/dispatch-protocol.js +0 -71
  41. package/src/utils/dispatch.js +0 -172
  42. package/src/utils/executor.js +0 -293
  43. package/src/utils/learnings-io.js +0 -76
  44. package/src/utils/learnings.js +0 -204
  45. package/src/utils/orchestrator-io.js +0 -356
  46. package/src/utils/orchestrator.js +0 -590
  47. package/src/utils/pricing.js +0 -28
  48. package/src/utils/providers/claude-code.js +0 -43
  49. package/src/utils/skill-loader.js +0 -83
  50. package/src/utils/trace.js +0 -400
  51. package/src/utils/workflow-parser.js +0 -225
@@ -1,172 +0,0 @@
1
- /**
2
- * dispatch.js — Validation and resolution utilities for the Guild dispatch protocol.
3
- *
4
- * Provides functions to validate workflow step configurations, resolve agent
5
- * metadata from frontmatter, determine effective model tiers, and resolve
6
- * tiers to concrete model IDs.
7
- */
8
-
9
- import { existsSync, readFileSync } from 'fs';
10
- import { join } from 'path';
11
- import { parseFrontmatter } from './files.js';
12
- import { parseSkill } from './workflow-parser.js';
13
- import {
14
- MODEL_TIERS,
15
- FAILURE_STRATEGIES,
16
- DEFAULT_AGENT_TIERS,
17
- DEFAULT_MODEL_PROFILES,
18
- FALLBACK_CHAIN,
19
- } from './dispatch-protocol.js';
20
-
21
- /**
22
- * Validates a workflow step configuration object.
23
- * @param {object} config - Step config with role, intent, model-tier, etc.
24
- * @returns {string[]} Array of error messages (empty means valid)
25
- */
26
- export function validateStepConfig(config) {
27
- const errors = [];
28
-
29
- if (!config.role) {
30
- errors.push('Missing required field: role');
31
- }
32
-
33
- if (!config.intent) {
34
- errors.push('Missing required field: intent');
35
- }
36
-
37
- if (config['model-tier'] && !MODEL_TIERS.includes(config['model-tier'])) {
38
- errors.push(`Invalid model-tier: "${config['model-tier']}". Must be one of: ${MODEL_TIERS.join(', ')}`);
39
- }
40
-
41
- if (config['on-failure']) {
42
- const isGoto = config['on-failure'].startsWith('goto:');
43
- if (!FAILURE_STRATEGIES.includes(config['on-failure']) && !isGoto) {
44
- errors.push(`Invalid on-failure: "${config['on-failure']}". Must be one of: ${FAILURE_STRATEGIES.join(', ')}, or goto:<step-id>`);
45
- }
46
- }
47
-
48
- if (config['max-retries'] !== undefined) {
49
- const val = config['max-retries'];
50
- if (!Number.isInteger(val) || val < 1) {
51
- errors.push(`Invalid max-retries: ${val}. Must be a positive integer`);
52
- }
53
- }
54
-
55
- return errors;
56
- }
57
-
58
- /**
59
- * Reads agent metadata from the agent's markdown frontmatter.
60
- * @param {string} role - Agent role name (e.g., 'tech-lead')
61
- * @param {string} [projectRoot=process.cwd()] - Project root directory
62
- * @returns {{ name: string, role: string, defaultTier: string|undefined, [key: string]: unknown } | null}
63
- */
64
- export function resolveAgentMetadata(role, projectRoot = process.cwd()) {
65
- const agentPath = join(projectRoot, '.claude', 'agents', `${role}.md`);
66
-
67
- if (!existsSync(agentPath)) {
68
- return null;
69
- }
70
-
71
- const content = readFileSync(agentPath, 'utf8');
72
- const frontmatter = parseFrontmatter(content);
73
-
74
- return {
75
- ...frontmatter,
76
- role,
77
- defaultTier: frontmatter['default-tier'] || undefined,
78
- };
79
- }
80
-
81
- /**
82
- * Resolves the effective model tier for a workflow step using the precedence chain:
83
- * 1. stepConfig['model-tier'] (explicit in workflow step)
84
- * 2. agentMetadata.defaultTier (from agent frontmatter)
85
- * 3. DEFAULT_AGENT_TIERS[role] (hardcoded defaults)
86
- * 4. 'execution' (ultimate fallback)
87
- *
88
- * @param {object} stepConfig - Workflow step with role and optional model-tier
89
- * @param {object|null} [agentMetadata=null] - Agent metadata from resolveAgentMetadata
90
- * @returns {string} One of MODEL_TIERS values
91
- */
92
- export function resolveEffectiveTier(stepConfig, agentMetadata = null) {
93
- if (stepConfig['model-tier'] && MODEL_TIERS.includes(stepConfig['model-tier'])) {
94
- return stepConfig['model-tier'];
95
- }
96
-
97
- if (agentMetadata?.defaultTier && MODEL_TIERS.includes(agentMetadata.defaultTier)) {
98
- return agentMetadata.defaultTier;
99
- }
100
-
101
- const defaultTier = DEFAULT_AGENT_TIERS[stepConfig.role];
102
- if (defaultTier) {
103
- return defaultTier;
104
- }
105
-
106
- return 'execution';
107
- }
108
-
109
- /**
110
- * Resolves a model tier to a concrete model ID using a profile.
111
- * Applies the fallback chain if the tier is not found in the profile.
112
- *
113
- * @param {string} tier - One of MODEL_TIERS
114
- * @param {string|Record<string, string>} profile - Profile name ('max', 'pro') or custom mapping
115
- * @returns {string} Concrete model ID (e.g., 'claude-opus-4-6')
116
- * @throws {Error} If no model can be resolved after exhausting the fallback chain
117
- */
118
- export function resolveModel(tier, profile) {
119
- const profileMap = typeof profile === 'string'
120
- ? DEFAULT_MODEL_PROFILES[profile]
121
- : profile;
122
-
123
- if (!profileMap) {
124
- throw new Error(`Unknown profile: "${profile}". Available: ${Object.keys(DEFAULT_MODEL_PROFILES).join(', ')}`);
125
- }
126
-
127
- let currentTier = tier;
128
- const visited = new Set();
129
-
130
- while (currentTier) {
131
- if (visited.has(currentTier)) {
132
- break;
133
- }
134
- visited.add(currentTier);
135
-
136
- if (profileMap[currentTier]) {
137
- return profileMap[currentTier];
138
- }
139
-
140
- currentTier = FALLBACK_CHAIN[currentTier];
141
- }
142
-
143
- throw new Error(`Cannot resolve model for tier "${tier}": no model available in profile after fallback chain`);
144
- }
145
-
146
- /**
147
- * Extracts dispatch configuration from skill markdown content.
148
- * Precedence: workflow steps (frontmatter) > null (legacy prose).
149
- *
150
- * Dependency direction: dispatch.js imports from workflow-parser.js.
151
- * Do not reverse this — workflow-parser.js must not import from dispatch.js.
152
- *
153
- * @param {string} skillMarkdown - Raw SKILL.md content
154
- * @returns {{ source: 'workflow', steps: Array<object> } | { source: null }}
155
- * @throws {Error} If YAML frontmatter is malformed (propagated from parseSkill)
156
- */
157
- export function extractDispatchConfigs(skillMarkdown) {
158
- if (!skillMarkdown) {
159
- return { source: null };
160
- }
161
-
162
- const skill = parseSkill(skillMarkdown);
163
-
164
- if (skill.workflow && Array.isArray(skill.workflow.steps) && skill.workflow.steps.length > 0) {
165
- return {
166
- source: 'workflow',
167
- steps: skill.workflow.steps,
168
- };
169
- }
170
-
171
- return { source: null };
172
- }
@@ -1,293 +0,0 @@
1
- /**
2
- * executor.js — Execution loop for Guild workflow plans.
3
- *
4
- * Drives a plan to completion by iterating through steps, dispatching
5
- * agent steps to a provider function and system steps to local commands.
6
- * Supports parallel execution (v1.2) and delegation to sub-skills.
7
- */
8
-
9
- import { execFile } from 'child_process';
10
- import {
11
- advanceStep,
12
- getNextSteps,
13
- isPlanComplete,
14
- MAX_DELEGATION_DEPTH,
15
- createExecutionPlan,
16
- } from './orchestrator.js';
17
- import {
18
- buildStepContext,
19
- recordStepTrace,
20
- loadWorkflow,
21
- resolveStepDispatch,
22
- } from './orchestrator-io.js';
23
-
24
- const SYSTEM_STEP_TIMEOUT = 120_000; // 2 minutes
25
-
26
- /**
27
- * Promisified execFile wrapper that always resolves (never rejects).
28
- *
29
- * @param {string} cmd - Command to execute
30
- * @param {string[]} args - Arguments
31
- * @param {object} opts - execFile options
32
- * @returns {Promise<{ stdout: string, stderr: string, exitCode: number }>}
33
- */
34
- function execFileAsync(cmd, args, opts) {
35
- return new Promise((resolve) => {
36
- execFile(cmd, args, opts, (error, stdout, stderr) => {
37
- resolve({
38
- stdout: stdout || '',
39
- stderr: stderr || (error && error.message) || '',
40
- exitCode: error ? (typeof error.code === 'number' ? error.code : 1) : 0,
41
- });
42
- });
43
- });
44
- }
45
-
46
- /**
47
- * Executes a system step by running its commands or handling delegation.
48
- *
49
- * @param {object} step - System step definition
50
- * @param {object} [options={}] - Options
51
- * @param {string} [options.projectRoot=process.cwd()] - Working directory for commands
52
- * @returns {Promise<{ status: string, output: string }>}
53
- */
54
- async function executeSystemStep(step, options = {}) {
55
- const { projectRoot = process.cwd() } = options;
56
-
57
- if (step.commands && step.commands.length > 0) {
58
- const outputs = [];
59
- for (const cmd of step.commands) {
60
- // v1.1: simple split — commands with quoted args or shell features
61
- // are not supported. Use simple commands like "npm test".
62
- const [bin, ...args] = cmd.split(' ');
63
- const result = await execFileAsync(bin, args, {
64
- cwd: projectRoot,
65
- timeout: SYSTEM_STEP_TIMEOUT,
66
- });
67
-
68
- if (result.exitCode !== 0) {
69
- return {
70
- status: 'failed',
71
- output: result.stderr || result.stdout || `Command failed: ${cmd}`,
72
- };
73
- }
74
- outputs.push(result.stdout);
75
- }
76
- return { status: 'passed', output: outputs.join('\n') };
77
- }
78
-
79
- if (step.delegatesTo) {
80
- return { status: 'passed', output: `System step with delegation — handled by executeDelegation` };
81
- }
82
-
83
- return { status: 'passed', output: 'System step completed' };
84
- }
85
-
86
- /**
87
- * Finds a step definition by ID across all groups in a plan.
88
- *
89
- * @param {object} plan - Execution plan
90
- * @param {string} stepId - Step ID to find
91
- * @returns {object|null}
92
- */
93
- function findStepInPlan(plan, stepId) {
94
- for (const group of plan.groups) {
95
- for (const step of group.steps) {
96
- if (step.id === stepId) return step;
97
- }
98
- }
99
- return null;
100
- }
101
-
102
- /**
103
- * Dispatches a single step (agent or system) and returns its result.
104
- *
105
- * @param {object} step - Step definition
106
- * @param {object} dispatch - Dispatch info for this step
107
- * @param {object} context - Execution context
108
- * @param {import('./orchestrator.js').ExecutionPlan} context.currentPlan - Current plan state
109
- * @param {Function} context.provider - Agent step provider
110
- * @param {string} context.projectRoot - Working directory
111
- * @param {string} context.skillBody - Skill body text
112
- * @param {object} context.executeOptions - Full options passed to execute()
113
- * @returns {Promise<{ status: string, output: string, outcome?: object, error?: string }>}
114
- */
115
- async function dispatchStep(step, dispatch, context) {
116
- const { currentPlan, provider, projectRoot, skillBody, executeOptions } = context;
117
-
118
- if (step.role === 'system' && step.delegatesTo) {
119
- return executeDelegation(step, executeOptions);
120
- }
121
-
122
- if (step.role === 'system') {
123
- return executeSystemStep(step, { projectRoot });
124
- }
125
-
126
- const stepContext = buildStepContext(step, currentPlan, { skillBody });
127
- return provider(step, dispatch, stepContext);
128
- }
129
-
130
- /**
131
- * Executes a delegation step by loading and running the sub-skill.
132
- *
133
- * @param {object} step - Delegation step (with delegatesTo field)
134
- * @param {object} options - Execute options from parent
135
- * @returns {Promise<{ status: string, output: string, error?: string }>}
136
- */
137
- async function executeDelegation(step, options) {
138
- const {
139
- provider,
140
- trace,
141
- projectRoot,
142
- profile = 'max',
143
- onStepStart,
144
- onStepEnd,
145
- delegationDepth = 0,
146
- } = options;
147
-
148
- if (delegationDepth >= MAX_DELEGATION_DEPTH) {
149
- return {
150
- status: 'failed',
151
- output: '',
152
- error: `Delegation depth limit (${MAX_DELEGATION_DEPTH}) exceeded at step "${step.id}" delegating to "${step.delegatesTo}"`,
153
- };
154
- }
155
-
156
- let subSkill;
157
- try {
158
- subSkill = loadWorkflow(step.delegatesTo);
159
- } catch (err) {
160
- return {
161
- status: 'failed',
162
- output: '',
163
- error: `Failed to load delegated skill "${step.delegatesTo}": ${err.message}`,
164
- };
165
- }
166
-
167
- const subPlan = createExecutionPlan(subSkill.workflow, {
168
- skillName: subSkill.name || step.delegatesTo,
169
- });
170
-
171
- const subDispatchMap = {};
172
- for (const group of subPlan.groups) {
173
- for (const s of group.steps) {
174
- subDispatchMap[s.id] = resolveStepDispatch(s, { profile, projectRoot });
175
- }
176
- }
177
-
178
- const finalSubPlan = await execute(subPlan, subDispatchMap, {
179
- provider,
180
- trace,
181
- projectRoot,
182
- skillBody: subSkill.body || '',
183
- onStepStart,
184
- onStepEnd,
185
- delegationDepth: delegationDepth + 1,
186
- profile,
187
- });
188
-
189
- if (finalSubPlan.status === 'completed') {
190
- return { status: 'passed', output: `Delegation to "${step.delegatesTo}" completed` };
191
- }
192
-
193
- return {
194
- status: 'failed',
195
- output: '',
196
- error: `Delegated skill "${step.delegatesTo}" ended with status: ${finalSubPlan.status}`,
197
- };
198
- }
199
-
200
- /**
201
- * Executes a workflow plan to completion.
202
- *
203
- * Drives the orchestrator state machine by repeatedly calling getNextSteps,
204
- * dispatching each step (agent via provider, system via local commands),
205
- * and advancing the plan with the result. Parallel groups are dispatched
206
- * concurrently via Promise.all.
207
- *
208
- * @param {import('./orchestrator.js').ExecutionPlan} plan - Initial execution plan
209
- * @param {Object.<string, import('./orchestrator-io.js').StepDispatchInfo>} dispatchInfoMap - Dispatch info per step
210
- * @param {object} [options={}] - Options
211
- * @param {Function} options.provider - Agent step provider: (step, dispatch, context) => { status, output, outcome?, error?, tokens? }
212
- * @param {object} [options.trace] - Trace context for recording step executions
213
- * @param {string} [options.projectRoot] - Working directory for system commands
214
- * @param {string} [options.skillBody=''] - Skill body text for context building
215
- * @param {Function} [options.onStepStart] - Callback before each step: (step, dispatch) => void
216
- * @param {Function} [options.onStepEnd] - Callback after each step: (step, result) => void
217
- * @param {number} [options.delegationDepth=0] - Current delegation nesting depth
218
- * @param {string} [options.profile='max'] - Model profile for delegation dispatch
219
- * @returns {Promise<import('./orchestrator.js').ExecutionPlan>} Final plan state
220
- */
221
- export async function execute(plan, dispatchInfoMap, options = {}) {
222
- const {
223
- provider,
224
- trace,
225
- projectRoot,
226
- skillBody = '',
227
- onStepStart,
228
- onStepEnd,
229
- } = options;
230
-
231
- let currentPlan = plan;
232
- let emptyIterations = 0;
233
- const MAX_EMPTY_ITERATIONS = 100;
234
-
235
- while (!isPlanComplete(currentPlan)) {
236
- const { steps, skipped } = getNextSteps(currentPlan);
237
-
238
- for (const stepId of skipped) {
239
- currentPlan = advanceStep(currentPlan, stepId, { status: 'skipped' });
240
-
241
- if (trace) {
242
- const step = findStepInPlan(currentPlan, stepId);
243
- const dispatch = dispatchInfoMap[stepId] || {};
244
- if (step) {
245
- recordStepTrace(trace, step, currentPlan.stepStates[stepId], dispatch);
246
- }
247
- }
248
- }
249
-
250
- if (steps.length === 0) {
251
- if (isPlanComplete(currentPlan)) break;
252
- if (++emptyIterations > MAX_EMPTY_ITERATIONS) {
253
- currentPlan = { ...currentPlan, status: 'aborted' };
254
- break;
255
- }
256
- continue;
257
- }
258
- emptyIterations = 0;
259
-
260
- const dispatchContext = {
261
- currentPlan,
262
- provider,
263
- projectRoot,
264
- skillBody,
265
- executeOptions: options,
266
- };
267
-
268
- const settled = await Promise.all(
269
- steps.map(async (step) => {
270
- const dispatch = dispatchInfoMap[step.id] || {};
271
- onStepStart?.(step, dispatch);
272
- const result = await dispatchStep(step, dispatch, dispatchContext);
273
- return { step, dispatch, result };
274
- })
275
- );
276
-
277
- for (const { step, dispatch, result } of settled) {
278
- currentPlan = advanceStep(currentPlan, step.id, result);
279
-
280
- if (trace) {
281
- recordStepTrace(trace, step, currentPlan.stepStates[step.id], dispatch);
282
- }
283
-
284
- onStepEnd?.(step, result);
285
- }
286
- }
287
-
288
- if (currentPlan.status === 'running' && isPlanComplete(currentPlan)) {
289
- currentPlan = { ...currentPlan, status: 'completed' };
290
- }
291
-
292
- return currentPlan;
293
- }
@@ -1,76 +0,0 @@
1
- /**
2
- * learnings-io.js — File I/O operations for compound learnings.
3
- *
4
- * Read, write, init, exists, and delete operations for .claude/guild/learnings.md.
5
- * Separated from the pure functions in learnings.js following the trace.js pattern.
6
- *
7
- * NOTE: File locking for concurrent access is intentionally omitted.
8
- * Concurrent workflow execution is a v2 concern — current Guild workflows
9
- * are single-session and sequential.
10
- */
11
-
12
- import { readFileSync, writeFileSync, existsSync, unlinkSync, mkdirSync } from 'fs';
13
- import { dirname } from 'path';
14
- import { GUILD_LEARNINGS_PATH, renderEmptyLearnings } from './learnings.js';
15
-
16
- /**
17
- * Reads the learnings file from disk.
18
- * Returns the raw content as a string, or null if the file does not exist.
19
- * @param {string} [filePath] - Override path (default: GUILD_LEARNINGS_PATH)
20
- * @returns {string | null}
21
- */
22
- export function readLearnings(filePath) {
23
- const target = filePath || GUILD_LEARNINGS_PATH;
24
- if (!existsSync(target)) return null;
25
- return readFileSync(target, 'utf8');
26
- }
27
-
28
- /**
29
- * Writes content to the learnings file.
30
- * Creates parent directories if needed.
31
- * @param {string} content - Markdown content to write
32
- * @param {string} [filePath] - Override path (default: GUILD_LEARNINGS_PATH)
33
- */
34
- export function writeLearnings(content, filePath) {
35
- const target = filePath || GUILD_LEARNINGS_PATH;
36
- mkdirSync(dirname(target), { recursive: true });
37
- writeFileSync(target, content, 'utf8');
38
- }
39
-
40
- /**
41
- * Checks whether the learnings file exists on disk.
42
- * @param {string} [filePath] - Override path (default: GUILD_LEARNINGS_PATH)
43
- * @returns {boolean}
44
- */
45
- export function learningsExist(filePath) {
46
- const target = filePath || GUILD_LEARNINGS_PATH;
47
- return existsSync(target);
48
- }
49
-
50
- /**
51
- * Initializes the learnings file with empty scaffold content.
52
- * No-ops if the file already exists.
53
- * @param {string} [projectName='Project'] - Project name for the header
54
- * @param {string} [filePath] - Override path (default: GUILD_LEARNINGS_PATH)
55
- * @returns {{ created: boolean }}
56
- */
57
- export function initLearnings(projectName = 'Project', filePath) {
58
- const target = filePath || GUILD_LEARNINGS_PATH;
59
- if (existsSync(target)) return { created: false };
60
- mkdirSync(dirname(target), { recursive: true });
61
- writeFileSync(target, renderEmptyLearnings(projectName), 'utf8');
62
- return { created: true };
63
- }
64
-
65
- /**
66
- * Deletes the learnings file from disk.
67
- * Returns { deleted: false } if the file does not exist (no throw).
68
- * @param {string} [filePath] - Override path (default: GUILD_LEARNINGS_PATH)
69
- * @returns {{ deleted: boolean }}
70
- */
71
- export function deleteLearnings(filePath) {
72
- const target = filePath || GUILD_LEARNINGS_PATH;
73
- if (!existsSync(target)) return { deleted: false };
74
- unlinkSync(target);
75
- return { deleted: true };
76
- }