feat(skills): add skill-local safe_mode and resource classification #161

New issue

Open

opened 2026-11-09 16:21:35 +00:00 by CoreRasurae · 0 comments

CoreRasurae commented

2026-11-09 16:21:35 +00:00

Member

Metadata

Commit Message: feat(skills): add skill-local safe_mode and resource classification
Branch: feature/m1-skill-safe-mode-resource-classification

Background and context

ADR-2034 D-8 and D-9 extend the Skill Package Schema. The current
cleveractors.agents.skill_schema module validates the original fields and a
generic resources mapping but knows nothing about a skill-local safety switch
or about the standard directory classes. Without this foundation the loader and
the skill tool cannot enforce the new policy.

Current behavior

Skill has no safe_mode field.
SkillValidator accepts any well-formed relative resource path and does not
classify it.
SkillLoader stores only a flat path-to-resource mapping.
Existing skill packages without the new field must remain valid and keep
their Package ID.

Expected behavior

Add an optional top-level boolean safe_mode with default true.
A non-boolean safe_mode raises the existing registry ValidationError.
Absence is the canonical representation of true; safe_mode: false is
retained through validation and is part of the content-addressed identity.
Classify every resource path by its first POSIX segment (case-sensitive):
references/, assets/, scripts/, non-standard top-level directory, or
root-level file.
Expose the classification through the validated value objects and the loaded
skill catalogue so the skill tool can enforce D-8/D-10 without re-parsing
paths.
No script execution or other privileged behaviour is added by this issue.

Acceptance criteria

A skill document omitting safe_mode validates with effective safe_mode: true and an unchanged Package ID.
A skill document with safe_mode: false validates and produces a different
Package ID from the same document without the field.
A skill document with a non-boolean safe_mode fails validation.
references/REFERENCE.md classifies as reference, assets/logo.png as asset,
scripts/run.py as script, fixtures/sample.pdf as non-standard, and
LICENSE.txt as root file.
Directory classification is case-sensitive: Scripts/run.py is non-standard.
Existing skill-resolution, schema, and loader tests remain green.
New Behave scenarios cover the field and every class; a Robot integration
scenario verifies the loaded catalogue exposes the class and safety state.
Coverage remains >= 97% via nox -s coverage_report.

Supporting information

ADR-2034 D-2 (schema), D-3 (identity), D-8 (taxonomy), D-9 (safe_mode)
Agent Skills specification: https://agentskills.io/specification
Modules: cleveractors.agents.skill_schema, cleveractors.agents.skills
Existing tests: features/agent_skills.feature,
features/steps/agent_skills_steps.py, robot/agent_skills.robot

Subtasks

Add safe_mode to Skill and SkillValidator with canonical absence
Add resource-path classification value objects/helpers
Carry classification and safe_mode through SkillLoader's catalogue
Tests (Behave): add scenarios for safe_mode validation and each resource class
Tests (Robot): verify the loaded catalogue exposes class and safety state
Verify coverage >=97% via nox -s coverage_report
Run nox (all default sessions), fix any errors

Definition of Done

This issue is complete when:

All subtasks above are completed and checked off.
A Git commit is created where the first line matches the Commit Message in
Metadata exactly.
The commit is pushed to the branch matching Branch in Metadata exactly.
The commit is submitted as a pull request to master, reviewed, and merged.

## Metadata - **Commit Message**: `feat(skills): add skill-local safe_mode and resource classification` - **Branch**: `feature/m1-skill-safe-mode-resource-classification` ## Background and context ADR-2034 D-8 and D-9 extend the Skill Package Schema. The current `cleveractors.agents.skill_schema` module validates the original fields and a generic `resources` mapping but knows nothing about a skill-local safety switch or about the standard directory classes. Without this foundation the loader and the skill tool cannot enforce the new policy. ## Current behavior - `Skill` has no `safe_mode` field. - `SkillValidator` accepts any well-formed relative resource path and does not classify it. - `SkillLoader` stores only a flat path-to-resource mapping. - Existing skill packages without the new field must remain valid and keep their Package ID. ## Expected behavior - Add an optional top-level boolean `safe_mode` with default `true`. - A non-boolean `safe_mode` raises the existing registry `ValidationError`. - Absence is the canonical representation of `true`; `safe_mode: false` is retained through validation and is part of the content-addressed identity. - Classify every resource path by its first POSIX segment (case-sensitive): `references/`, `assets/`, `scripts/`, non-standard top-level directory, or root-level file. - Expose the classification through the validated value objects and the loaded skill catalogue so the skill tool can enforce D-8/D-10 without re-parsing paths. - No script execution or other privileged behaviour is added by this issue. ## Acceptance criteria - A skill document omitting `safe_mode` validates with effective `safe_mode: true` and an unchanged Package ID. - A skill document with `safe_mode: false` validates and produces a different Package ID from the same document without the field. - A skill document with a non-boolean `safe_mode` fails validation. - `references/REFERENCE.md` classifies as reference, `assets/logo.png` as asset, `scripts/run.py` as script, `fixtures/sample.pdf` as non-standard, and `LICENSE.txt` as root file. - Directory classification is case-sensitive: `Scripts/run.py` is non-standard. - Existing skill-resolution, schema, and loader tests remain green. - New Behave scenarios cover the field and every class; a Robot integration scenario verifies the loaded catalogue exposes the class and safety state. - Coverage remains >= 97% via `nox -s coverage_report`. ## Supporting information - ADR-2034 D-2 (schema), D-3 (identity), D-8 (taxonomy), D-9 (`safe_mode`) - Agent Skills specification: https://agentskills.io/specification - Modules: `cleveractors.agents.skill_schema`, `cleveractors.agents.skills` - Existing tests: `features/agent_skills.feature`, `features/steps/agent_skills_steps.py`, `robot/agent_skills.robot` ## Subtasks - [ ] Add `safe_mode` to `Skill` and `SkillValidator` with canonical absence - [ ] Add resource-path classification value objects/helpers - [ ] Carry classification and `safe_mode` through `SkillLoader`'s catalogue - [ ] Tests (Behave): add scenarios for `safe_mode` validation and each resource class - [ ] Tests (Robot): verify the loaded catalogue exposes class and safety state - [ ] Verify coverage >=97% via `nox -s coverage_report` - [ ] Run `nox` (all default sessions), fix any errors ## Definition of Done This issue is complete when: - All subtasks above are completed and checked off. - A Git commit is created where the first line matches the Commit Message in Metadata exactly. - The commit is pushed to the branch matching Branch in Metadata exactly. - The commit is submitted as a pull request to `master`, reviewed, and merged.

Rows
Columns