[Feature]: Add DeepSeek Harness configured-asset inventory for Skills and Instructions

[Cobb04]

Preflight

• I searched existing issues for a similar request.

What problem are you trying to solve?

Better Harness can analyze persisted DeepSeek Harness (dsh) session evidence and DSH can discover and explicitly invoke the canonical Better Harness Skill, but agent-customize still cannot answer:

What capabilities and instructions is this DSH environment configured with?

Specifically, it cannot inventory the effective filesystem-configured DSH Skills or the DSH Instruction sources applicable to a selected working directory.

This Story concerns configuration evidence only. A configured Skill does not prove it was invoked, and an applicable Instruction file does not prove it influenced a particular session.

What would success look like?

Better Harness can produce a bounded DSH configured-assets inventory for a selected workspace and cwd through one canonical DSH provider:

• DSH is registered for AGENT_CUSTOMIZE.
• Effective filesystem Skill winners appear in manage.skills using native DSH roots, precedence, validation, and the supported <root>/<name>/SKILL.md and <root>/<name>.md layouts. Invalid, skipped, and shadowed candidates are excluded from active Skills and may be reported through diagnostics.
• Applicable Instructions appear in manage.rules in native order: the authorized global source followed by the project-root → cwd chain, with native base candidates such as AGENTS.md and CLAUDE.md and local overlays such as AGENTS.local.md and CLAUDE.local.md.
• Native same-directory duplicate-content behavior and bounded per-source/aggregate byte behavior are preserved without emitting Skill bodies or Instruction prose.
• Inventory uses an explicit cwd when supplied and otherwise uses the workspace; it does not derive cwd from session evidence.
• Project sources are available by default. User/global sources are neither probed nor read without explicit opt-in, while explicitly supplied external roots authorize only those named roots.
• The qualified DSH filesystem defaults and explicit overrides are supported, with honest diagnostics for unresolved Cordis, Profile, Preset, and runtime/in-process Skill state.
• A pinned, credential-free native DSH verification covers the important filesystem, ordering, shadowing, budgeting, authorization, symlink, and cross-platform path boundaries.

Runtime/in-process Skill providers remain outside the completeness claim.

Where would this be used?

The primary surface is Better Harness agent-customize with provider dsh, producing configured-asset evidence for an analyzed DSH workspace/cwd.

This does not establish Public Quickstart readiness, complete DSH host support, lifecycle support, report-loop support, or runtime use of the reported assets.

Examples or evidence

• Issue #93, PR #94, PR #97, and PR #98 established and hardened the earlier DSH session-evidence slice.
• Issue #99 and PR #100 established verified native Better Harness Skill discovery and explicit invocation; [Feature]: Add verified DeepSeek Harness Skill discovery with explicit-only invocation #99 explicitly left configured-assets inventory out of scope.
• DSH's native filesystem Skill provider has DSH-specific roots, layouts, precedence, and winner selection.
• DSH's native Instruction configuration, file discovery, and rendering show that the applicable chain is cwd-sensitive, ordered, deduplicated, and byte-bounded.

Anything else?

Use the existing configured-assets model: effective filesystem Skills map to manage.skills, ordered Instructions map to manage.rules, and skipped, shadowed, malformed, or unresolved state uses existing diagnostics/unsupported conventions. One provider at scripts/agent-customize/providers/dsh.mjs should own both collections. This Story should add AGENT_CUSTOMIZE, not ASSET_PRACTICES, and configured-assets logic should remain separate from the DSH session-analysis adapter.

Out of scope are runtime/plugin/scoped Skill enumeration; complete Cordis, Profile, or Preset resolution; MCP, Plugin, or Profile inventory; lifecycle and session-analysis changes; ASSET_PRACTICES; practice reports, baselines, lint, evidence-bundle integration, and report/output routing; Public Quickstart; full DSH parity; upstream DSH changes; a new shared schema; and any claim that configured assets were used at runtime.

DSH remains a developer preview, so the support claim should stay pinned and qualified.

Is this bounded filesystem Configured Assets contract — native Skills plus cwd-sensitive Instructions, with qualified defaults/explicit overrides and honest diagnostics for unresolved Cordis/Profile/Preset runtime state — the right first AGENT_CUSTOMIZE slice for DSH?

I'd be happy to take this implementation after scope alignment.

[phodal]

Thanks for the detailed proposal. This looks like the right bounded first AGENT_CUSTOMIZE slice for DSH. Please go ahead and open a PR.