Helm 0.5.0

Helm 0.5.0

Helm 0.5.0 extends skill-local harness governance into skill-document governance.

The previous release made manifest quality operator-auditable. This release raises the bar again: a skill is no longer treated as operationally strong just because contract.json looks narrow. When SKILL.md exists, Helm now expects it to expose the workflow's real operating contracts as well.

Highlights

• expanded Helm's skill quality model so SKILL.md is treated as an operator-facing contract
• refreshed docs/skill-quality-and-policy.md around explicit input, decision, output, and failure contracts
• updated references/skill-capture-template.md so newly created skills start from contract-oriented structure instead of descriptive prose
• extended python3 scripts/run_with_profile.py audit-manifest-quality --json to inspect SKILL.md structure and basic manifest-to-document consistency
• repositioned README and operator guidance so Helm is clearly framed as a runtime-agnostic skill governance layer, not a skill catalog

Recommended validation flow

python3 -m pip install --user --no-build-isolation .
helm --help
python3 scripts/run_with_profile.py validate-manifests --json
python3 scripts/run_with_profile.py audit-manifest-quality --json
helm survey --path examples/demo-workspace
helm validate --path examples/demo-workspace

Notes

• This release does not try to ship a larger public skill library.
• The main change is that Helm now treats weak skill prose as an operational quality problem, not just a documentation problem.
• A good skill in Helm is now expected to expose both execution policy in contract.json and operating procedure in SKILL.md.

Helm 0.4.0

Helm 0.4.0

Helm 0.4.0 turns manifest-based harness governance into an operator-facing quality loop.

The previous release moved execution policy into skill-local manifests. This release makes that structure auditable: operators can now detect contracts that still look like generic backfills, then tighten them before treating the workspace as stable.

Highlights

• added python3 scripts/run_with_profile.py audit-manifest-quality --json
• tightened representative skill contracts so profile scope, context hydration, and approval boundaries are skill-specific rather than generic
• refreshed README and operator docs around smaller-model stability, skill-owned contracts, and policy maintenance
• clarified that extensibility should come from per-skill contracts and audit loops, not from central harness edits

Recommended validation flow

python3 -m pip install --user --no-build-isolation .
helm --help
python3 scripts/run_with_profile.py validate-manifests --json
python3 scripts/run_with_profile.py audit-manifest-quality --json
helm survey --path examples/demo-workspace
helm validate --path examples/demo-workspace
helm harness --path examples/demo-workspace contract --skill travel-ops-ko

Notes

• This release is about operational precision and auditability, not a claim that model-quality limits disappeared.
• The recommended release bar is now both structural validity and contract-quality validity.
• New skills should still start narrow and only widen after the actual workflow proves it is necessary.

Helm 0.3.0

Helm 0.3.0

Helm 0.3.0 moves the adaptive harness from a central skill registry to skill-local manifests.

The release focus is operational scalability. New skills no longer require central harness edits just to declare execution policy, default profile, or strict runner expectations.

Highlights

• replaced the central references/skill_contracts.json registry with per-skill skills/<skill>/contract.json
• moved allowed_profiles and default_profile into the skill manifest itself
• added python3 scripts/run_with_profile.py validate-manifests --json for missing or malformed manifest detection
• updated helm validate so workspace validation includes manifest audit results
• refreshed release docs and README around manifest-based harness governance

Recommended validation flow

python3 -m pip install --user --no-build-isolation .
helm --help
python3 scripts/run_with_profile.py validate-manifests --json
helm survey --path examples/demo-workspace
helm validate --path examples/demo-workspace
helm harness --path examples/demo-workspace contract --skill travel-ops-ko

Notes

• This release improves extensibility and reduces central hardcoding. It does not claim to solve model-quality limits.
• The default skill scaffold now starts from inspect_local only. Wider permissions should be added intentionally per skill.
• Existing workspaces can continue to keep legacy references/skill_profile_policies.json as fallback, but new skill governance should live in the manifest.

Helm 0.2.0

Helm 0.2.0

Helm 0.2.0 extends the initial public release with explicit task finalization inspection.

The main change is that durable capture planning is no longer only buried inside the task ledger. Operators can now inspect recent finalization state, pending durable capture follow-up, and checkpoint-linked finalization context directly from the CLI.

Highlights

• added helm context recent-state for recent finalized task inspection
• added helm memory pending-captures for the current durable capture queue
• added helm ops capture-state for finalization count summaries
• added helm checkpoint finalize to inspect a task's finalization plan together with rollback context
• refreshed release docs and README examples around finalization-driven operations

Recommended validation flow

python3 -m pip install --user --no-build-isolation .
helm --help
helm memory --help
helm survey --path examples/demo-workspace
helm context --path examples/demo-workspace recent-state --limit 5
helm ops --path examples/demo-workspace capture-state --limit 10

Notes

• Helm still treats workspace-specific durable memory mutation as runtime-specific.
• The new commands improve visibility and operator review; they do not try to overwrite private runtime memory stores by themselves.
• This release is the natural follow-up to 0.1.0, not a packaging-only refresh.

Helm 0.1.0

Helm 0.1.0

Initial public release of Helm as a stability-first operations layer for long-lived personal agents.

Highlights

• packaged helm CLI with one-line install flow
• dedicated Helm workspace model using .helm/
• guided onboarding for OpenClaw, Hermes, and Markdown note sources
• file-native context hydration across notes, memory, ontology, tasks, commands, and checkpoints
• checkpoint recommendation and rollback guidance
• draft skill review, approval, and rejection workflow
• example demo workspace and clearer public documentation

Recommended first-run flow

curl -fsSL https://raw.githubusercontent.com/JDeun/Helm/main/install.sh | bash
helm onboard --path ~/.helm/workspace --use-detected

Notes

• Helm does not overwrite an existing OpenClaw or Hermes tree by default.
• Obsidian is optional, but adopting an existing Markdown vault as a read-only source is recommended when available.
• This release is intended as the first public, usable release of the extracted Helm operations layer.