Add DeepSeek Harness (DSH) integration

[AnInteger]

Adds a skills-based integration for DeepSeek Harness (DSH) — the agent runtime behind the dsh CLI and its Web GUI. Closes #4334.

How DSH consumes Spec Kit

DSH discovers skills one level deep from .dsh/skills (its native project root, highest provider rank) and .agents/skills, using speckit-<name>/SKILL.md directory bundles with name/description frontmatter — the same agentskills.io layout SkillsIntegration already scaffolds. Users invoke skills through the /-trigger input (e.g. /speckit-specify <feature description>); the harness injects the rendered skill content into the turn. Project AGENTS.md guidance is loaded natively by DSH, so no context-file handling is needed in the integration.

What changed

Area	Change
src/specify_cli/integrations/dsh/	New SkillsIntegration subclass: installs to .dsh/skills/, slash-style invocations, CLI dispatch via dsh --profile headless "<prompt>", multi_install_safe (isolated .dsh/ root coexists with other agents)
src/specify_cli/integrations/__init__.py	Registration (import + _register(), alphabetical)
src/specify_cli/_invocation_style.py	dsh added to ALWAYS_SLASH_AGENTS so init output and hook messages render /speckit-<command>
src/specify_cli/commands/init.py	Next-steps hint for dsh skills mode
integrations/catalog.json	Catalog entry (keeps TestCatalogParity green)
.github/ISSUE_TEMPLATE/*	Supported-agent lists/dropdowns (kept in sync per test_issue_template_agent_lists_match_runtime_integrations)
docs/reference/integrations.md	Docs table row
tests/integrations/test_integration_dsh.py	40 tests: inherited SkillsIntegrationTests suite, build_exec_args regression, DSH-discovery compatibility invariants (kebab-case names, name+description frontmatter, one-level-deep bundles), multi-install isolation, hook invocation rendering

No templates/commands/*.md files are touched, so no existing command behavior changes.

Test selection reasoning

Changed file	Affects	Test	Why
src/specify_cli/integrations/dsh/__init__.py	specify init --integration dsh	T1	New integration subpackage — scaffolding path
src/specify_cli/integrations/__init__.py	specify init / integration install	T1	Registry registration wires the new key
src/specify_cli/_invocation_style.py	init next-steps display, hook invocation rendering	T1	dsh added to ALWAYS_SLASH_AGENTS
src/specify_cli/commands/init.py	specify init output	T1	New next-steps hint
catalog / issue templates / docs / tests	none (metadata/docs/tests)	—	No runtime behavior

Manual test results

Agent: DeepSeek Harness (dsh CLI v0.1.1-rc.2, headless profile) | OS/Shell: Linux (WSL2) / bash

Command tested	Notes
specify init --integration dsh --script sh (T1)	PASS — 10 skills installed to .dsh/skills/speckit-<name>/SKILL.md; next-steps panel shows the dsh web hint and /speckit-<command> usage; integration list shows dsh (CLI required: yes, multi-install safe: yes); integration uninstall dsh removed all 10 tracked files cleanly
/speckit-constitution (T2)	PASS — executed by a real DSH agent via dsh --profile headless "/speckit-constitution …"; the skill loaded through the /-token gesture boundary; the agent created .specify/memory/constitution.md with a sync impact report and self-validated; exit 0
/speckit-specify (T3)	PASS — executed by a real DSH agent; create-new-feature.sh ran and created specs/001-markdown-notes-cli/; generated spec.md (user stories with priorities, Given/When/Then acceptance scenarios, 11 functional requirements) and checklists/requirements.md; the agent ran the requirements checklist and reported all items passing; exit 0

Additional evidence:

• DSH provider smoke test — instantiated the real @deepseek-ai/dsh-skill-filesystem FileSystemSkillProvider against a scaffolded project: all 10 speckit-* skills discovered from the project-dsh root with invocation: {modelInvocable: true, userInvocable: true}; full skill bodies loadable through provider.get().
• Automated tests — .venv/bin/python -m pytest tests -q: 7231 passed / 156 skipped. The 25 failures are pre-existing PowerShell-encoding environment failures, reproduced identically on a clean main checkout of this machine (verified via git stash), unrelated to this change.

AI disclosure

This PR was implemented with AI assistance (DeepSeek Harness, model: glm-5.3) under human supervision: the integration approach and each design decision (target directory, multi-install safety, CLI dispatch, invocation style) were presented for review and approved by the filer before submission, and all changes and manual test evidence were personally reviewed and re-verified. The commit carries an Assisted-by: trailer to the same effect.