feat(knowledge): treat informational invocations as safe (1.3.0)

[gmpassos]

Context

Many commands are risky-by-default in the knowledge base — dart, flutter, and the interpreters in shell_knowledge.dart (node, python, ruby, java, …) plus cargo/go carry baseCapabilities: {executePrograms}. The analysis pipeline only ever adds capabilities and raises risk, so harmless invocations like dart --version, node --version, python --version were wrongly reported as executePrograms.

This adds a new mechanism — an informational-form gate — that recognizes purely informational invocations as read-only/safe, only when every argument is informational (so dart can still run arbitrary code with other parameters).

What changed

• New global defaultInformationalTokens (--version, --help, --usage) applied to all known commands.
• New per-command CommandKnowledge.informationalTokens field for tool-specific forms: java -version, python -V, node/php/ruby -v, perl -v/-V, deno/cargo -V, go version/env, dotnet --info/--list-sdks/--list-runtimes.
• Gate runs at the top of _collect, before base caps/risk and at every recursion depth → wrapped forms like sudo node --version are handled (and sudo still reports privilegeEscalation).
• Case-sensitive matching (-V ≠ -v); a single payload token defeats the gate (dart --version x.dart, go env -w K=V still report executePrograms).

Behavior change

Bare --version/--help/--usage (and the per-command short forms) on execute-by-default tools now classify as read-only/safe instead of executePrograms. npm --version moves from an empty capability set to readFilesystem (risk stays safe).

Verification

• dart format — clean
• dart analyze — No issues found
• dart test — 404/404 pass (13 new gate tests)

Bumps version to 1.3.0 and updates CHANGELOG.md + doc/architecture.md.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!