@bradtaylorsf/alpha-loop 1.4.2 → 1.5.0

package/README.md

@@ -48,17 +48,23 @@ Alpha Loop implements a 12-step pipeline for each issue:
 1. **Status Update** — Labels issue `in-progress`, assigns to you, updates project board
 2. **Worktree** — Creates an isolated git worktree so work doesn't conflict with other issues
 3. **Plan** — Agent analyzes the issue and enriches it with implementation details
-4. **Implement** — Agent writes the code, guided by project vision, context, and learnings from previous issues
+3b. **Fetch Comments** — Loads issue comments so the agent has the full conversation context
+4. **Implement** — Agent writes the code, guided by project vision, context, comments, and learnings from previous issues
 5. **Test + Retry** — Runs your test command; if tests fail, agent fixes and retries (up to `max_test_retries`)
 6. **Verify + Retry** — Starts your dev server, uses playwright-cli to test the feature like a real user, takes screenshots
 7. **Review** — A review agent reads the diff, checks for gaps, security issues, and missing wiring — fixes what it can
 8. **Create PR** — Opens a PR with test results, review summary, and verification status
+8b. **Assumptions** — Agent summarizes assumptions and decisions made, posts as a comment on the issue for user validation
 9. **Learn** — Extracts learnings (patterns, anti-patterns, what worked/failed) for future sessions
 10. **Update Issue** — Posts results as a comment, updates labels
 11. **Auto-Merge** — Merges the PR to the session branch (if enabled)
 12. **Cleanup** — Removes the worktree
 
-After all issues are processed, Alpha Loop generates a **session summary** that aggregates learnings across issues and produces actionable recommendations.
+After all issues are processed, Alpha Loop:
+1. **Auto-captures failures** as eval cases for regression testing
+2. Generates a **session summary** aggregating learnings across issues
+3. Runs a **post-session code review** on the full session diff to catch cross-issue integration problems
+4. Creates the **session PR** with all findings included
 
 ### Milestone-Based Workflow
 
@@ -101,6 +107,65 @@ Run `alpha-loop review` to trigger the self-improvement loop. It reads all accum
 
 Without `--apply`, proposals are saved to `learnings/proposed-updates/` for review. With `--apply`, changes are written and a draft PR is created.
 
+### Eval System (`alpha-loop eval`)
+
+Alpha Loop includes a self-improving eval system inspired by [Meta-Harness](https://arxiv.org/abs/2603.28052) (Lee et al., 2026). It captures real failures as eval cases and tracks composite scores over time to measure whether prompt/skill changes actually help.
+
+```bash
+# Capture failures from recent sessions as eval cases
+alpha-loop eval capture
+
+# Run the eval suite and compute composite score
+alpha-loop eval run
+
+# View score history, Pareto frontier, or compare runs
+alpha-loop eval scores
+alpha-loop eval pareto
+alpha-loop eval compare 1 2
+
+# Greedy search over model configurations per pipeline step
+alpha-loop eval search --models "haiku,sonnet,opus"
+
+# Estimate cost before running
+alpha-loop eval estimate
+
+# Compare two config files side-by-side
+alpha-loop eval compare-configs config-a.yaml config-b.yaml
+```
+
+Eval cases live in `.alpha-loop/evals/` and scores are appended to `scores.jsonl` (Git-friendly, append-only). The composite score formula is pass-rate primary with lightweight penalties for retries and duration. Real API costs (tokens, USD) are tracked per case from agent output and used for the Pareto frontier.
+
+Step-level evals test individual pipeline stages (plan, implement, test, test-fix, review, learn, skill) and run in seconds using LLM-judge and keyword checks:
+
+```bash
+# Run only step-level evals (fast, cheap)
+alpha-loop eval --suite step
+
+# Run evals for a specific step
+alpha-loop eval --suite step --step review
+
+# Convert between AlphaLoop and skill-creator eval formats
+alpha-loop eval convert --direction to-skill
+alpha-loop eval convert --direction from-skill --input path/to/evals.json
+
+# Import SWE-bench cases from HuggingFace (requires Python + datasets)
+alpha-loop eval import-swebench --count 10 --repo "django/django"
+```
+
+### Evolve (`alpha-loop evolve`)
+
+The evolve command runs a Meta-Harness-style optimization loop: a proposer agent reads full execution traces, scores, and source code, then proposes targeted changes to prompts, skills, or config. Changes are evaluated against the eval suite — improvements are kept, regressions are reverted (autoresearch keep/discard pattern).
+
+```bash
+alpha-loop evolve                         # Run up to 5 iterations
+alpha-loop evolve --max-iterations 10     # Run 10 iterations
+alpha-loop evolve --continuous            # Run until manually stopped (Ctrl-C)
+alpha-loop evolve --surface prompts       # Only modify agent prompts (safest)
+alpha-loop evolve --surface all           # Modify prompts + pipeline code (riskier)
+alpha-loop evolve --resume                # Resume from a previous evolve session
+alpha-loop evolve --dry-run               # Preview without changes
+```
+
 ### Crash Recovery (`alpha-loop resume`)
 
 If the loop hangs or crashes mid-session, work can be stranded on local branches with no PR. Run `alpha-loop resume` to recover:
@@ -124,7 +189,7 @@ During live verification, the agent takes screenshots at key states and saves th
 | `alpha-loop run` | Fetch matching issues, process them all, then exit |
 | `alpha-loop run --dry-run` | Preview without making changes |
 | `alpha-loop scan` | Generate/refresh project context and instructions file |
-| `alpha-loop vision` | Interactive project vision setup (`.alpha-loop/vision.md`) |
+| `alpha-loop vision` | **(deprecated)** Use `alpha-loop plan` instead |
 | `alpha-loop auth` | Save authenticated browser state for verification |
 | `alpha-loop history` | View session history |
 | `alpha-loop history <name>` | View a specific session |
@@ -135,6 +200,28 @@ During live verification, the agent takes screenshots at key states and saves th
 | `alpha-loop resume --issue <N>` | Resume a specific issue |
 | `alpha-loop review` | Analyze learnings and propose self-improvements |
 | `alpha-loop review --apply` | Apply proposed improvements and create a draft PR |
+| `alpha-loop eval` | Run the eval suite and compute composite score |
+| `alpha-loop eval capture` | Capture failures as eval cases (interactive) |
+| `alpha-loop eval list` | Show eval cases and recent scores |
+| `alpha-loop eval scores` | Show score history over time |
+| `alpha-loop eval pareto` | Show score/cost Pareto frontier |
+| `alpha-loop eval compare <r1> <r2>` | Compare two eval runs |
+| `alpha-loop eval search` | Greedy search over model configurations per pipeline step |
+| `alpha-loop eval estimate` | Estimate cost of running the eval suite |
+| `alpha-loop eval compare-configs <a> <b>` | Compare two YAML config files side-by-side |
+| `alpha-loop eval convert` | Convert between AlphaLoop and skill-creator eval formats |
+| `alpha-loop eval import-swebench` | Import eval cases from SWE-bench dataset |
+| `alpha-loop evolve` | Meta-Harness-style automated optimization loop |
+| `alpha-loop plan` | Generate a full project scope (milestones + issues) from seed inputs using AI |
+| `alpha-loop plan --seed <file>` | Read seed description from a file instead of prompting |
+| `alpha-loop plan --dry-run` | Display the plan without creating any GitHub resources |
+| `alpha-loop plan --yes --seed <file>` | Non-interactive mode: accept all AI recommendations |
+| `alpha-loop triage` | Analyze and improve existing issues (staleness, clarity, size, duplicates, support ticket enrichment) |
+| `alpha-loop triage --dry-run` | Display findings without making changes |
+| `alpha-loop triage --yes` | Non-interactive mode: apply all AI-recommended triage actions |
+| `alpha-loop roadmap` | Organize open issues into milestones using AI analysis |
+| `alpha-loop roadmap --dry-run` | Display proposed roadmap without making changes |
+| `alpha-loop roadmap --yes` | Non-interactive mode: apply all AI-recommended assignments |
 
 ### Run Options
 
@@ -176,6 +263,15 @@ harnesses:
 # Safety limits (0 = unlimited)
 max_issues: 20
 max_session_duration: 7200  # 2 hours in seconds
+
+# Post-session review (runs after all issues, reviews full session diff)
+post_session:
+  review: true
+  security_scan: true
+
+# Eval system
+auto_capture: true  # capture failures as eval cases
+eval_dir: .alpha-loop/evals
 ```
 
 ### Configuration Reference
@@ -210,6 +306,14 @@ max_session_duration: 7200  # 2 hours in seconds
 | `run_full` | `false` | Run full pipeline without skipping any steps |
 | `verbose` | `false` | Enable verbose agent output |
 | `harnesses` | (auto from agent) | Coding harnesses to sync skills/agents to (e.g., `claude`, `codex`) |
+| `eval_dir` | `.alpha-loop/evals` | Directory for eval cases and scores |
+| `eval_model` | (agent default) | AI model for eval judging |
+| `eval_timeout` | `300` | Timeout in seconds for eval case execution |
+| `auto_capture` | `true` | Auto-capture failures as eval cases at end of session |
+| `pipeline` | `{}` | Per-step agent/model overrides (see below) |
+| `pricing` | (built-in) | Custom token pricing per model for cost tracking |
+| `post_session.review` | `true` | Run holistic code review on full session diff |
+| `post_session.security_scan` | `true` | Include security scanning in post-session review |
 
 ### Environment Variables
 
@@ -243,6 +347,12 @@ All config options can be set via environment variables (uppercase, same names):
 | `MERGE_TO` | `merge_to` |
 | `RUN_FULL` | `run_full` |
 | `VERBOSE` | `verbose` |
+| `EVAL_DIR` | `eval_dir` |
+| `EVAL_MODEL` | `eval_model` |
+| `EVAL_TIMEOUT` | `eval_timeout` |
+| `AUTO_CAPTURE` | `auto_capture` |
+| `SKIP_POST_SESSION_REVIEW` | `post_session.review` (inverted) |
+| `SKIP_POST_SESSION_SECURITY` | `post_session.security_scan` (inverted) |
 
 **Precedence:** CLI flags > environment variables > `.alpha-loop.yaml` > auto-detection > defaults
 
@@ -278,6 +388,27 @@ harnesses:
   - claude  # also sync to Claude for teammates using it
 ```
 
+### Per-Step Pipeline Config
+
+Use `pipeline` to assign different models to different pipeline stages. This lets you use cheaper models for simple steps and reserve expensive models for implementation:
+
+```yaml
+agent: claude
+model: claude-sonnet-4-6  # default for all steps
+
+pipeline:
+  plan:
+    model: claude-haiku-4-5       # cheap model for planning
+  implement:
+    model: claude-sonnet-4-6      # main model for coding
+  review:
+    model: claude-opus-4-6        # best model for review
+  learn:
+    model: claude-haiku-4-5       # cheap model for learning
+```
+
+Use `alpha-loop eval search` to automatically find the best model assignment per step via greedy coordinate descent over your eval suite.
+
 ## GitHub Setup
 
 ### Labels
@@ -331,6 +462,8 @@ What needs to be done.
 | `.alpha-loop/vision.md` | Yes | Project vision document |
 | `.alpha-loop/context.md` | Yes | Auto-generated project context |
 | `.alpha-loop/learnings/` | Yes | Learning files, session manifests, and session summaries (shared with team) |
+| `.alpha-loop/evals/` | Yes | Eval cases (YAML) and score history (`scores.jsonl`) |
+| `.alpha-loop/traces/` | No (gitignored) | Meta-Harness style execution traces per session |
 | `.alpha-loop/sessions/` | No (gitignored) | Local session logs, results JSON, screenshots |
 | `.alpha-loop/auth/` | No (gitignored) | Saved browser auth state for verification |
 | `.worktrees/` | No (gitignored) | Temporary git worktrees during processing |

package/dist/cli.js

@@ -8,7 +8,7 @@ import { syncCommand } from './commands/sync.js';
 program
     .name('alpha-loop')
     .description('Agent-agnostic automated development loop')
-    .version('1.4.2');
+    .version('1.5.0');
 program
     .command('init')
     .description('Full project onboarding: config, templates, vision, scan, sync')
@@ -47,7 +47,7 @@ program
     .action(scanCommand);
 program
     .command('vision')
-    .description('Interactive project vision setup')
+    .description('(deprecated) Interactive project vision setup — use "plan" instead')
     .action(visionCommand);
 program
     .command('auth')
@@ -58,6 +58,35 @@ program
     .description('Sync .alpha-loop/templates/ to all configured harnesses')
     .option('--check', 'Check for drift without syncing (exits non-zero if drift found)')
     .action(syncCommand);
+program
+    .command('plan')
+    .description('Generate a full project scope (milestones + issues) from seed inputs using AI')
+    .option('--seed <file>', 'Read seed description from a file instead of prompting')
+    .option('--no-vision', 'Skip vision generation even if no vision exists')
+    .option('--dry-run', 'Display the plan without creating any GitHub resources')
+    .option('-y, --yes', 'Skip interactive prompts, accept all AI recommendations')
+    .action(async (options) => {
+    const { planCommand } = await import('./commands/plan.js');
+    await planCommand(options);
+});
+program
+    .command('triage')
+    .description('Analyze and improve existing issues (staleness, clarity, size, duplicates)')
+    .option('--dry-run', 'Display findings without making changes')
+    .option('-y, --yes', 'Skip interactive prompts, accept all AI recommendations')
+    .action(async (options) => {
+    const { triageCommand } = await import('./commands/triage.js');
+    await triageCommand(options);
+});
+program
+    .command('roadmap')
+    .description('Organize open issues into milestones using AI analysis')
+    .option('--dry-run', 'Display proposed roadmap without making changes')
+    .option('-y, --yes', 'Skip interactive prompts, accept all AI recommendations')
+    .action(async (options) => {
+    const { roadmapCommand } = await import('./commands/roadmap.js');
+    await roadmapCommand(options);
+});
 program
     .command('resume')
     .description('Resume stranded work — push branches, run review, open PRs')
@@ -87,7 +116,7 @@ evalCmd
     .option('--suite <suite>', 'Run only a suite: step (fast) or e2e (slow)')
     .option('--case <id>', 'Run a single eval case by ID prefix')
     .option('--type <type>', 'Filter by type: full or step')
-    .option('--step <step>', 'Filter by pipeline step (plan, implement, test, review, verify)')
+    .option('--step <step>', 'Filter by pipeline step (plan, implement, test, test-fix, review, verify, learn, skill)')
     .option('--verbose', 'Show detailed output')
     .action(async (options) => {
     const { evalRunCommand } = await import('./commands/eval.js');
@@ -116,10 +145,14 @@ evalCmd
 });
 evalCmd
     .command('search')
-    .description('Greedy search over model/agent configurations')
+    .description('Greedy coordinate descent search over model/agent configurations')
     .option('--models <models>', 'Models to test (comma-separated)')
     .option('--agents <agents>', 'Agents to test (comma-separated)')
     .option('--max-runs <n>', 'Maximum number of eval runs')
+    .option('--budget <n>', 'Maximum number of eval runs (alias for --max-runs)')
+    .option('--step <step>', 'Only search over this pipeline step')
+    .option('--min-score <score>', 'Minimum acceptable score threshold')
+    .option('--optimize <target>', 'Optimize for: cost or efficiency (default: efficiency)')
     .action(async (options) => {
     const { evalSearchCommand } = await import('./commands/eval.js');
     await evalSearchCommand(options);
@@ -138,17 +171,51 @@ evalCmd
     const { evalCompareCommand } = await import('./commands/eval.js');
     evalCompareCommand(run1, run2);
 });
+evalCmd
+    .command('estimate')
+    .description('Estimate cost of running the eval suite with current or specified config')
+    .option('--config <path>', 'Path to a YAML config file to estimate')
+    .action(async (options) => {
+    const { evalEstimateCommand } = await import('./commands/eval.js');
+    evalEstimateCommand(options);
+});
+evalCmd
+    .command('compare-configs <configA> <configB>')
+    .description('Compare two YAML config files side-by-side')
+    .action(async (configA, configB) => {
+    const { evalCompareConfigsCommand } = await import('./commands/eval.js');
+    evalCompareConfigsCommand(configA, configB);
+});
 evalCmd
     .command('import-swebench')
     .description('Import eval cases from SWE-bench dataset')
-    .action(async () => {
+    .option('--dataset <path>', 'Path to a downloaded JSONL file (skips auto-download)')
+    .option('--dataset-id <id>', 'HuggingFace dataset ID (default: princeton-nlp/SWE-bench_Lite)')
+    .option('--count <n>', 'Maximum number of cases to import')
+    .option('--repo <owner/repo>', 'Filter by repository (e.g. django/django)')
+    .option('--ids <csv>', 'Import specific instance IDs (comma-separated)')
+    .option('--step <step>', 'Pipeline step to target (default: implement)')
+    .action(async (options) => {
     const { evalImportSwebenchCommand } = await import('./commands/eval.js');
-    await evalImportSwebenchCommand();
+    await evalImportSwebenchCommand(options);
+});
+evalCmd
+    .command('convert')
+    .description('Convert between AlphaLoop eval format and skill-creator format')
+    .option('--direction <dir>', 'Conversion direction: to-skill or from-skill (default: to-skill)')
+    .option('--input <path>', 'Input file path (for from-skill)')
+    .option('--output <path>', 'Output file path')
+    .action(async (options) => {
+    const { evalConvertCommand } = await import('./commands/eval.js');
+    evalConvertCommand(options);
 });
 program
     .command('evolve')
     .description('Meta-Harness-style automated optimization loop')
     .option('--max-iterations <n>', 'Maximum optimization iterations (default: 5)')
+    .option('--continuous', 'Run until manually stopped (SIGINT)')
+    .option('--surface <level>', 'Optimization surface: prompts, skills, config, all (default: prompts)')
+    .option('--resume', 'Resume from a previous evolve session')
     .option('--dry-run', 'Preview without making changes')
     .option('--verbose', 'Show detailed agent output')
     .action(async (options) => {

package/dist/cli.js.map

@@ -1 +1 @@
-{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAEjD,OAAO;KACJ,IAAI,CAAC,YAAY,CAAC;KAClB,WAAW,CAAC,2CAA2C,CAAC;KACxD,OAAO,CAAC,OAAO,CAAC,CAAC;AAEpB,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,gEAAgE,CAAC;KAC7E,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,WAAW,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC3D,MAAM,WAAW,EAAE,CAAC;AACtB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,KAAK,CAAC;KACd,WAAW,CAAC,2EAA2E,CAAC;KACxF,MAAM,CAAC,WAAW,EAAE,yBAAyB,CAAC;KAC9C,MAAM,CAAC,iBAAiB,EAAE,iBAAiB,CAAC;KAC5C,MAAM,CAAC,cAAc,EAAE,qBAAqB,CAAC;KAC7C,MAAM,CAAC,eAAe,EAAE,kBAAkB,CAAC;KAC3C,MAAM,CAAC,cAAc,EAAE,0BAA0B,CAAC;KAClD,MAAM,CAAC,oBAAoB,EAAE,uCAAuC,CAAC;KACrE,MAAM,CAAC,cAAc,EAAE,kCAAkC,CAAC;KAC1D,MAAM,CAAC,qBAAqB,EAAE,wDAAwD,CAAC;KACvF,MAAM,CAAC,QAAQ,EAAE,4BAA4B,CAAC;KAC9C,MAAM,CAAC,WAAW,EAAE,sCAAsC,CAAC;KAC3D,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,IAAI,OAAO,CAAC,IAAI;QAAE,OAAO,CAAC,SAAS,GAAG,CAAC,CAAC;IACxC,MAAM,UAAU,CAAC,OAAO,CAAC,CAAC;AAC5B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,mBAAmB,CAAC;KAC5B,WAAW,CAAC,sBAAsB,CAAC;KACnC,MAAM,CAAC,MAAM,EAAE,+BAA+B,CAAC;KAC/C,MAAM,CAAC,SAAS,EAAE,yBAAyB,CAAC;KAC5C,MAAM,CAAC,cAAc,CAAC,CAAC;AAE1B,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,kCAAkC,CAAC;KAC/C,MAAM,CAAC,WAAW,CAAC,CAAC;AAEvB,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,kCAAkC,CAAC;KAC/C,MAAM,CAAC,aAAa,CAAC,CAAC;AAEzB,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,kCAAkC,CAAC;KAC/C,MAAM,CAAC,WAAW,CAAC,CAAC;AAEvB,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,yDAAyD,CAAC;KACtE,MAAM,CAAC,SAAS,EAAE,iEAAiE,CAAC;KACpF,MAAM,CAAC,WAAW,CAAC,CAAC;AAEvB,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,4DAA4D,CAAC;KACzE,MAAM,CAAC,eAAe,EAAE,qCAAqC,CAAC;KAC9D,MAAM,CAAC,kBAAkB,EAAE,0CAA0C,CAAC;KACtE,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,CAAC;IAC/D,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,2FAA2F,CAAC;KACxG,MAAM,CAAC,SAAS,EAAE,4CAA4C,CAAC;KAC/D,MAAM,CAAC,kBAAkB,EAAE,gDAAgD,CAAC;KAC5E,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,CAAC;IAC/D,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,mBAAmB;AACnB,MAAM,OAAO,GAAG,OAAO;KACpB,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,2DAA2D,CAAC,CAAC;AAE5E,OAAO;KACJ,OAAO,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;KACnC,WAAW,CAAC,gDAAgD,CAAC;KAC7D,MAAM,CAAC,eAAe,EAAE,kCAAkC,CAAC;KAC3D,MAAM,CAAC,iBAAiB,EAAE,6CAA6C,CAAC;KACxE,MAAM,CAAC,aAAa,EAAE,qCAAqC,CAAC;KAC5D,MAAM,CAAC,eAAe,EAAE,8BAA8B,CAAC;KACvD,MAAM,CAAC,eAAe,EAAE,iEAAiE,CAAC;KAC1F,MAAM,CAAC,WAAW,EAAE,sBAAsB,CAAC;KAC3C,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC9D,MAAM,cAAc,CAAC,OAAO,CAAC,CAAC;AAChC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,iBAAiB,CAAC;KAC1B,WAAW,CAAC,8CAA8C,CAAC;KAC3D,MAAM,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;IACtB,MAAM,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAClE,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC;AACtC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,mCAAmC,CAAC;KAChD,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC/D,eAAe,EAAE,CAAC;AACpB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,8BAA8B,CAAC;KAC3C,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACjE,iBAAiB,EAAE,CAAC;AACtB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,+CAA+C,CAAC;KAC5D,MAAM,CAAC,mBAAmB,EAAE,kCAAkC,CAAC;KAC/D,MAAM,CAAC,mBAAmB,EAAE,kCAAkC,CAAC;KAC/D,MAAM,CAAC,gBAAgB,EAAE,6BAA6B,CAAC;KACvD,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACjE,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;AACnC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,iCAAiC,CAAC;KAC9C,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACjE,iBAAiB,EAAE,CAAC;AACtB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,uBAAuB,CAAC;KAChC,WAAW,CAAC,gDAAgD,CAAC;KAC7D,MAAM,CAAC,KAAK,EAAE,IAAY,EAAE,IAAY,EAAE,EAAE;IAC3C,MAAM,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAClE,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;AACjC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,iBAAiB,CAAC;KAC1B,WAAW,CAAC,0CAA0C,CAAC;KACvD,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,yBAAyB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACzE,MAAM,yBAAyB,EAAE,CAAC;AACpC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,gDAAgD,CAAC;KAC7D,MAAM,CAAC,sBAAsB,EAAE,8CAA8C,CAAC;KAC9E,MAAM,CAAC,WAAW,EAAE,gCAAgC,CAAC;KACrD,MAAM,CAAC,WAAW,EAAE,4BAA4B,CAAC;KACjD,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,CAAC;IAC/D,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,OAAO,CAAC,KAAK,EAAE,CAAC"}
+{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAEjD,OAAO;KACJ,IAAI,CAAC,YAAY,CAAC;KAClB,WAAW,CAAC,2CAA2C,CAAC;KACxD,OAAO,CAAC,OAAO,CAAC,CAAC;AAEpB,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,gEAAgE,CAAC;KAC7E,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,WAAW,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC3D,MAAM,WAAW,EAAE,CAAC;AACtB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,KAAK,CAAC;KACd,WAAW,CAAC,2EAA2E,CAAC;KACxF,MAAM,CAAC,WAAW,EAAE,yBAAyB,CAAC;KAC9C,MAAM,CAAC,iBAAiB,EAAE,iBAAiB,CAAC;KAC5C,MAAM,CAAC,cAAc,EAAE,qBAAqB,CAAC;KAC7C,MAAM,CAAC,eAAe,EAAE,kBAAkB,CAAC;KAC3C,MAAM,CAAC,cAAc,EAAE,0BAA0B,CAAC;KAClD,MAAM,CAAC,oBAAoB,EAAE,uCAAuC,CAAC;KACrE,MAAM,CAAC,cAAc,EAAE,kCAAkC,CAAC;KAC1D,MAAM,CAAC,qBAAqB,EAAE,wDAAwD,CAAC;KACvF,MAAM,CAAC,QAAQ,EAAE,4BAA4B,CAAC;KAC9C,MAAM,CAAC,WAAW,EAAE,sCAAsC,CAAC;KAC3D,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,IAAI,OAAO,CAAC,IAAI;QAAE,OAAO,CAAC,SAAS,GAAG,CAAC,CAAC;IACxC,MAAM,UAAU,CAAC,OAAO,CAAC,CAAC;AAC5B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,mBAAmB,CAAC;KAC5B,WAAW,CAAC,sBAAsB,CAAC;KACnC,MAAM,CAAC,MAAM,EAAE,+BAA+B,CAAC;KAC/C,MAAM,CAAC,SAAS,EAAE,yBAAyB,CAAC;KAC5C,MAAM,CAAC,cAAc,CAAC,CAAC;AAE1B,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,kCAAkC,CAAC;KAC/C,MAAM,CAAC,WAAW,CAAC,CAAC;AAEvB,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,oEAAoE,CAAC;KACjF,MAAM,CAAC,aAAa,CAAC,CAAC;AAEzB,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,kCAAkC,CAAC;KAC/C,MAAM,CAAC,WAAW,CAAC,CAAC;AAEvB,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,yDAAyD,CAAC;KACtE,MAAM,CAAC,SAAS,EAAE,iEAAiE,CAAC;KACpF,MAAM,CAAC,WAAW,CAAC,CAAC;AAEvB,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,+EAA+E,CAAC;KAC5F,MAAM,CAAC,eAAe,EAAE,wDAAwD,CAAC;KACjF,MAAM,CAAC,aAAa,EAAE,iDAAiD,CAAC;KACxE,MAAM,CAAC,WAAW,EAAE,wDAAwD,CAAC;KAC7E,MAAM,CAAC,WAAW,EAAE,yDAAyD,CAAC;KAC9E,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,WAAW,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC3D,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;AAC7B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,4EAA4E,CAAC;KACzF,MAAM,CAAC,WAAW,EAAE,yCAAyC,CAAC;KAC9D,MAAM,CAAC,WAAW,EAAE,yDAAyD,CAAC;KAC9E,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,CAAC;IAC/D,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,SAAS,CAAC;KAClB,WAAW,CAAC,wDAAwD,CAAC;KACrE,MAAM,CAAC,WAAW,EAAE,iDAAiD,CAAC;KACtE,MAAM,CAAC,WAAW,EAAE,yDAAyD,CAAC;KAC9E,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;IACjE,MAAM,cAAc,CAAC,OAAO,CAAC,CAAC;AAChC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,4DAA4D,CAAC;KACzE,MAAM,CAAC,eAAe,EAAE,qCAAqC,CAAC;KAC9D,MAAM,CAAC,kBAAkB,EAAE,0CAA0C,CAAC;KACtE,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,CAAC;IAC/D,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,2FAA2F,CAAC;KACxG,MAAM,CAAC,SAAS,EAAE,4CAA4C,CAAC;KAC/D,MAAM,CAAC,kBAAkB,EAAE,gDAAgD,CAAC;KAC5E,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,CAAC;IAC/D,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,mBAAmB;AACnB,MAAM,OAAO,GAAG,OAAO;KACpB,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,2DAA2D,CAAC,CAAC;AAE5E,OAAO;KACJ,OAAO,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;KACnC,WAAW,CAAC,gDAAgD,CAAC;KAC7D,MAAM,CAAC,eAAe,EAAE,kCAAkC,CAAC;KAC3D,MAAM,CAAC,iBAAiB,EAAE,6CAA6C,CAAC;KACxE,MAAM,CAAC,aAAa,EAAE,qCAAqC,CAAC;KAC5D,MAAM,CAAC,eAAe,EAAE,8BAA8B,CAAC;KACvD,MAAM,CAAC,eAAe,EAAE,yFAAyF,CAAC;KAClH,MAAM,CAAC,WAAW,EAAE,sBAAsB,CAAC;KAC3C,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC9D,MAAM,cAAc,CAAC,OAAO,CAAC,CAAC;AAChC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,iBAAiB,CAAC;KAC1B,WAAW,CAAC,8CAA8C,CAAC;KAC3D,MAAM,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;IACtB,MAAM,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAClE,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC;AACtC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,WAAW,CAAC,mCAAmC,CAAC;KAChD,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC/D,eAAe,EAAE,CAAC;AACpB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,8BAA8B,CAAC;KAC3C,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACjE,iBAAiB,EAAE,CAAC;AACtB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,kEAAkE,CAAC;KAC/E,MAAM,CAAC,mBAAmB,EAAE,kCAAkC,CAAC;KAC/D,MAAM,CAAC,mBAAmB,EAAE,kCAAkC,CAAC;KAC/D,MAAM,CAAC,gBAAgB,EAAE,6BAA6B,CAAC;KACvD,MAAM,CAAC,cAAc,EAAE,oDAAoD,CAAC;KAC5E,MAAM,CAAC,eAAe,EAAE,qCAAqC,CAAC;KAC9D,MAAM,CAAC,qBAAqB,EAAE,oCAAoC,CAAC;KACnE,MAAM,CAAC,qBAAqB,EAAE,wDAAwD,CAAC;KACvF,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACjE,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;AACnC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,iCAAiC,CAAC;KAC9C,MAAM,CAAC,KAAK,IAAI,EAAE;IACjB,MAAM,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACjE,iBAAiB,EAAE,CAAC;AACtB,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,uBAAuB,CAAC;KAChC,WAAW,CAAC,gDAAgD,CAAC;KAC7D,MAAM,CAAC,KAAK,EAAE,IAAY,EAAE,IAAY,EAAE,EAAE;IAC3C,MAAM,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAClE,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;AACjC,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,UAAU,CAAC;KACnB,WAAW,CAAC,0EAA0E,CAAC;KACvF,MAAM,CAAC,iBAAiB,EAAE,wCAAwC,CAAC;KACnE,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,mBAAmB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACnE,mBAAmB,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,qCAAqC,CAAC;KAC9C,WAAW,CAAC,4CAA4C,CAAC;KACzD,MAAM,CAAC,KAAK,EAAE,OAAe,EAAE,OAAe,EAAE,EAAE;IACjD,MAAM,EAAE,yBAAyB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACzE,yBAAyB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;AAC9C,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,iBAAiB,CAAC;KAC1B,WAAW,CAAC,0CAA0C,CAAC;KACvD,MAAM,CAAC,kBAAkB,EAAE,uDAAuD,CAAC;KACnF,MAAM,CAAC,mBAAmB,EAAE,gEAAgE,CAAC;KAC7F,MAAM,CAAC,aAAa,EAAE,mCAAmC,CAAC;KAC1D,MAAM,CAAC,qBAAqB,EAAE,2CAA2C,CAAC;KAC1E,MAAM,CAAC,aAAa,EAAE,gDAAgD,CAAC;KACvE,MAAM,CAAC,eAAe,EAAE,8CAA8C,CAAC;KACvE,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,yBAAyB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IACzE,MAAM,yBAAyB,CAAC,OAAO,CAAC,CAAC;AAC3C,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,SAAS,CAAC;KAClB,WAAW,CAAC,gEAAgE,CAAC;KAC7E,MAAM,CAAC,mBAAmB,EAAE,kEAAkE,CAAC;KAC/F,MAAM,CAAC,gBAAgB,EAAE,kCAAkC,CAAC;KAC5D,MAAM,CAAC,iBAAiB,EAAE,kBAAkB,CAAC;KAC7C,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAClE,kBAAkB,CAAC,OAAO,CAAC,CAAC;AAC9B,CAAC,CAAC,CAAC;AAEL,OAAO;KACJ,OAAO,CAAC,QAAQ,CAAC;KACjB,WAAW,CAAC,gDAAgD,CAAC;KAC7D,MAAM,CAAC,sBAAsB,EAAE,8CAA8C,CAAC;KAC9E,MAAM,CAAC,cAAc,EAAE,qCAAqC,CAAC;KAC7D,MAAM,CAAC,mBAAmB,EAAE,uEAAuE,CAAC;KACpG,MAAM,CAAC,UAAU,EAAE,uCAAuC,CAAC;KAC3D,MAAM,CAAC,WAAW,EAAE,gCAAgC,CAAC;KACrD,MAAM,CAAC,WAAW,EAAE,4BAA4B,CAAC;KACjD,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;IACxB,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,CAAC;IAC/D,MAAM,aAAa,CAAC,OAAO,CAAC,CAAC;AAC/B,CAAC,CAAC,CAAC;AAEL,OAAO,CAAC,KAAK,EAAE,CAAC"}

package/dist/commands/eval.d.ts

@@ -13,6 +13,14 @@ export type EvalSearchOptions = {
     models?: string;
     agents?: string;
     maxRuns?: string;
+    /** Only search over this pipeline step. */
+    step?: string;
+    /** Minimum acceptable score. */
+    minScore?: string;
+    /** What to optimize: 'cost' or 'efficiency'. */
+    optimize?: string;
+    /** Maximum number of eval runs (alias for maxRuns). */
+    budget?: string;
 };
 /**
  * Run the eval suite.
@@ -40,14 +48,61 @@ export declare function evalListCommand(): void;
  */
 export declare function evalScoresCommand(): void;
 /**
- * Show score/cost Pareto frontier.
+ * Show score/cost Pareto frontier with ASCII chart.
  */
 export declare function evalParetoCommand(): void;
 /**
- * Greedy search over model/agent configs.
+ * Greedy coordinate descent search over model/agent configs.
+ *
+ * Strategy:
+ *   1. Establish baseline: run eval with current config → S₀, C₀
+ *   2. For each pipeline step, try alternative models (holding others fixed)
+ *   3. Keep Pareto-optimal changes (better score at ≤ cost, or same score at lower cost)
+ *   4. Repeat until no step can be improved or budget is exhausted
  */
 export declare function evalSearchCommand(options: EvalSearchOptions): Promise<void>;
+export type EvalImportSwebenchOptions = {
+    dataset?: string;
+    datasetId?: string;
+    count?: string;
+    repo?: string;
+    ids?: string;
+    step?: string;
+};
+/**
+ * Import SWE-bench cases from HuggingFace or a local JSONL file.
+ *
+ * Downloads entries from HuggingFace (requires Python + datasets library),
+ * converts each to a directory-based eval case under .alpha-loop/evals/cases/e2e/,
+ * and updates config.yaml with repo base commit mappings.
+ */
+export declare function evalImportSwebenchCommand(options?: EvalImportSwebenchOptions): Promise<void>;
+export type EvalConvertOptions = {
+    direction?: string;
+    input?: string;
+    output?: string;
+};
+/**
+ * Convert between AlphaLoop eval format and skill-creator format.
+ *
+ * Directions:
+ *   to-skill     — Convert AlphaLoop eval case → skill-creator evals.json
+ *   from-skill   — Convert skill-creator evals.json → AlphaLoop eval cases
+ */
+export declare function evalConvertCommand(options: EvalConvertOptions): void;
+export type EvalEstimateOptions = {
+    config?: string;
+};
+/**
+ * Estimate cost of running the eval suite with a given config.
+ * Shows per-step breakdown using pricing table and average token estimates.
+ */
+export declare function evalEstimateCommand(options: EvalEstimateOptions): void;
+export type EvalCompareConfigsOptions = {
+    configA: string;
+    configB: string;
+};
 /**
- * Import SWE-bench cases (placeholder — requires HuggingFace download).
+ * Compare two YAML config files side-by-side showing per-step model/agent differences.
  */
-export declare function evalImportSwebenchCommand(): Promise<void>;
+export declare function evalCompareConfigsCommand(configAPath: string, configBPath: string): void;