[konard]

Adds command-stream/core so shell-only consumers can use $, ProcessRunner and virtual commands without loading compatibility layers, terminals or development tools. Core loads 70 ESM / 71 CommonJS modules, compared with 284 for the full entry.

Fixes #235

Changes

• Share the existing shell implementation, built-in commands, registrations, settings and runner identity between the full entry, typed ESM/CommonJS core entries and process-runner.
• Separate virtual-command composition and metadata from the package installer that imports Execa.
• Enforce fewer than 100 cold-loaded non-builtin modules in both core formats. Reject compatibility, terminal, REPL/debugger and testing modules in the graph; bundle and execute core without those inputs.
• Document the public core API and prepare its minor release with a changeset.
• Update Execa to 10.2.1 after its strict TypeScript blocker closed. Update fancy-regex and both Rust lockfiles to fix dependency freshness checks, with a Rust patch fragment.
• Retry Deno imports only when esm.sh reports HTTP 408 while a cold build is queued: at most three attempts, with 5- and 10-second delays. Preserve live stderr and a bounded diagnostic tail; persistent timeouts, compiler errors and runtime failures remain fatal. Both real HTTP probes remain required.

The issue explicitly allows the lightweight-subpath approach. The full API remains available; installed dependency size is unchanged. The parity-exempt label applies to JavaScript's module-loader graph, which has no Rust counterpart. Executable language parity checks remain enabled.

Reproduction and regression coverage

Run node experiments/issue-235/import-probe.mjs command-stream or substitute command-stream/core. Before implementation, the full entry loaded 284 modules in approximately 461 ms on Linux / Node 26.12.1. Commit 7bb8fe2 preserves four failing core regressions before the subpath existed. Module count is the deterministic CI budget; elapsed time is diagnostic.

core-entry.test.mjs checks fresh ESM/CommonJS loads, the module budget, executable bundling, real commands, synchronous and exact-argv execution, virtual pipelines, shared state and API identity. Strict TypeScript consumers verify both declaration formats.

The previous Deno failure belongs to 285f44b, created 2026-10-24T08:20:11Z. Downloaded job-log line 550 reports an esm.sh Execa import returning HTTP 408; line 565 records exit code 1. The same dependency import and immutable GitHub-head URL failed locally, and the CDN status endpoint confirmed queued builds. The queue subsequently recovered and both real probes passed locally. Ten mocked regression tests verify retry success, exhaustion, ANSI diagnostics and immediate failure for other errors. No manual action is required.

Validation

• Full Bun suite: 2,812 passed, 10 skipped, zero failures across 179 files, including all ten CDN retry regressions.
• Node core budget, bundling, execution and shared-state checks: 5 passed.
• Both real Deno CDN probes passed locally, exercising the HTTP-served checkout and immutable GitHub head, shell execution, ShellJS identity, normal brace globs and depth protection.
• Lint, formatting, duplication and strict TypeScript checks passed. The pre-commit hook also passed.
• Existing Node regression suite: 124 passed; zx: 490 passed, 2 skipped; Execa: 38 passed.
• Rust all-feature suite including doctests: 905 passed, 7 ignored. Formatting, Clippy and rustdoc with warnings denied, and generated documentation checks passed.
• Declared and installed dependency freshness checks passed. Package dry-run includes the core source files and both declaration formats.

CI on 4f88be8eb7d7fa93cd84410cbc1c2933de003f2c: 49 passed, zero failed or pending, with six release-only jobs correctly skipped on this PR. All eight workflows completed successfully:

• JavaScript checks, including the Deno CDN check, all Node/Bun platform suites and production-install modes.
• Rust checks and package validation, including Linux, macOS and Windows.
• Dependency freshness, repository quality, language parity, workflows, benchmarks and security.

Downloaded Deno job-log lines 605–608 and 625–628 confirm both real probes executed all assertions successfully. The first Windows Rust attempt stalled in the existing PID-identity test and was cancelled for log collection; the targeted rerun passed on the same commit, including that test and all doctests. Both attempts' logs are preserved locally. No checks were disabled or relaxed.

The latest default branch is included, the working tree is clean, and the complete diff has been reviewed. Implementation and verification are complete; this PR is ready for review and remains unmerged.

[konard]

Working session summary

Implemented and pushed to PR #236.

• Added command-stream/core: 70–71 loaded modules versus 284.
• Local tests passed: 2,800 Bun tests and 905 Rust tests.
• 48 CI checks passed; working tree is clean.

Completion is blocked by esm.sh returning HTTP 408 in the required Deno check, confirmed locally and on CI retry. Evidence is in the PR, which remains draft and unmerged.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Model: GPT-6.1 Sol
• Provider: OpenAI
• Public pricing estimate: $2.820997

📊 Context and tokens usage:

• 190.2K / 272K (70%) input tokens, 45.0K / 128K (35%) output tokens

Total: (190.2K + 6.9M cached) input tokens, 45.0K output tokens, $2.820997 cost

🤖 Models used:

• Tool: OpenAI Codex
• Requested: gpt-6.1-sol
• Thinking level: xhigh (~31999 tokens)
• Model: GPT-6.1 Sol (gpt-6.1-sol)

📎 Log file uploaded as Gist (1524KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🔄 Auto-restart 1/5

Reason: CI failures detected

Evidence:

• Failing check: Import HTTP package on Deno with esm.sh dependencies — https://github.com/link-foundation/command-stream/actions/runs/40077631780/job/120825515833

Starting a new session to address it.

[konard]

Working session summary

Fixed in PR #236: bounded retries for esm.sh HTTP 408 errors, covered by 10 regression tests.

• CI: 49 passed, six release-only jobs skipped.
• Local Bun tests: 2,812 passed.
• PR ready for review and unmerged; latest main included, working tree clean.

Elapsed: 36 minutes.

---

This summary was automatically extracted from the AI working session output.

[konard]

🔄 Auto-restart-until-mergeable Log 1/5

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Model: GPT-6.1 Sol
• Provider: OpenAI
• Public pricing estimate: $3.750726

📊 Context and tokens usage:

• 252.4K / 272K (93%) input tokens, 51.4K / 128K (40%) output tokens

Total: (252.4K + 9.8M cached) input tokens, 51.4K output tokens, $3.750726 cost

🤖 Models used:

• Tool: OpenAI Codex
• Requested: gpt-6.1-sol
• Thinking level: xhigh (~31999 tokens)
• Model: GPT-6.1 Sol (gpt-6.1-sol)

📎 Log file uploaded as Gist (3071KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🎉 Auto-merged

This pull request has been automatically merged by hive-mind.

• All CI checks have passed

Linked issues

• 🔒 Closed by GitHub: import { $ } loads 281 modules (ShellJS, execa, zx, Bun shell, terminal capture) and takes ~350 ms #235