Skip to content

CHANGELOG.md

Latest commit

 

History

History
144 lines (112 loc) · 5.6 KB

CHANGELOG.md

File metadata and controls

144 lines (112 loc) · 5.6 KB
Raw

Changelog

Unreleased

Changed

  • Clarified that the documented upgrade path is npm i -g @to-agent/agent-exec@latest across README surfaces.
  • Narrowed aexec update to a convenience wrapper around the documented npm update plus version diagnostics. It no longer attempts package-manager or PATH repair.
  • aexec update --restart now restarts only when the active command appears to match npm latest.

0.2.3

Added

  • Added aexec update --restart so update can optionally restart the running server after package update.
  • Added restart pass-through options for update, including --public, --force, -f, --host, --port, and --use-project-env.

0.2.2

This release tightens ACL editing from the CLI and restores format-aware recovery for unknown non-API paths.

Added

  • Added aexec acl for listing, adding, removing, and checking exec.allow rules from the command line.
  • Added aexec acl remove --contains <text> for removing matching user exec.allow rules after confirmation.

Changed

  • Unknown non-API paths now return recovery guidance that follows the requested format. Markdown, HTML, JSON, and SJS requests point back to the matching root skill document surface.
  • JSON 404 recovery now includes the SJS recovery entry point while keeping the default response JSON.
  • ACL documentation now states that user ACL edits are picked up by a running server on the next /api/exec request.
  • Posting to SJS skill documents now returns an SJS fallback instead of a plain route miss. The /api/exec/SKILL.s.js fallback points back to the executable /api/exec surface and request body form.

Behavior Notes

  • aexec acl add writes to the primary user settings file.
  • Exact ACL edits ask for confirmation in a TTY and require --yes in non-interactive use.
  • Broad glob or regexp allow rules require --force --yes in non-interactive use.

0.2.1

This release adds a reset command for returning a local agent-exec installation to a fresh safe state without silently losing the previous config.

Added

  • Added aexec reset to back up the current local config directory and recreate a fresh minimal config.
  • Added aexec reset --keep-api-key for keeping the current API key while resetting settings and plugins.
  • Added aexec reset --api-key <key> for reproducible lab or device setup.
  • Added aexec reset --dry-run and aexec reset --json for inspection and automation.

Behavior Notes

  • Reset backs up by default to ~/.to-agent/backups/agent-exec/reset-*.
  • Fresh reset settings allow only aexec --version.
  • --no-backup is available only as an explicit destructive mode.

0.2.0

This release is written against the published @to-agent/agent-exec@0.1.10 package. It updates the agent-facing documentation surfaces, execution fallback responses, plugin generation defaults, and release packaging.

Added

  • Added Skill Script documentation surfaces through .s.js, .sjs, and Accept: text/sjs.
  • Added a shared SJS renderer in modules/sjs.js for root, index, API, skill, runtime, and fallback surfaces.
  • Added raw skill document variants such as SKILL.raw.json, SKILL.raw.html, SKILL.raw.s.js, and SKILL.raw.sjs for inspecting skill documents without ae: directive expansion.
  • Added ae: Markdown directives for projecting selected Markdown JSON blocks into generated JSON, HTML, and SJS metadata.
  • Added /skills/private/SKILL.md as the public entry document for the private skill namespace without listing private skill names.
  • Added this changelog.

Changed

  • Root and index documents now expose direct SJS variants while ordinary HTML and Markdown entry points remain available.
  • Public API skill documents remain discoverable, while API runtime calls, private routes, and CLI routes remain protected by API key checks.
  • /api/exec invalid-request responses now expose the expected request body form and point back to /api/exec/SKILL.s.js.
  • /api/exec SJS fallback responses are available when the request asks for text/sjs.
  • /api/exec continues to reject unexpected JSON body fields such as cmd, command, env, cwd, and shell; memo is now allowed as optional request metadata.
  • ACL and execution documentation now make the relationship between allowed command strings and args arrays more explicit.
  • Root, API, skills, private, and plugin skill documents were updated for the current discovery and execution surfaces.

Behavior Notes

  • Skill Script is directly accessible, but it is not required for ordinary human use of the HTTP interface.
  • Unknown non-API paths return recovery information without making SJS the default visible format.
  • Fresh installs remain conservative and do not generate broad wildcard command allow rules by default.

Plugin Generation And Safety

  • aexec plugin create no longer generates broad cmd * style allow rules by default.
  • aexec starterkit now generates ACL entries from detected help/version commands instead of broad wildcard execution rules.
  • Generated plugin skill files include request body metadata that can be used by JSON, HTML, and SJS outputs.
  • aexec plugin doctor warns about active broad wildcard allow rules.
  • Starterkit and plugin generation help output was normalized for direct script and command usage.

Internal Quality

  • Expanded unit and e2e coverage for SJS rendering, ae: directives, fallback responses, plugin generation, starterkit output, ACL behavior, private/public skill separation, and packaging.
  • Verified package contents include the new runtime SJS module and private namespace skill document.