feat(skills/agents): integrate sk harness quality gates across all workflow skills and agents

.github/agents/browse-ui-host-state.agent.md

@@ -95,3 +95,23 @@ Use Playwright or browser smoke only when runtime behavior cannot be proven by u
 ## Output
 
 Summarize the state machine, affected UI surfaces, and validation evidence. Call out any browser-dependent behavior explicitly.
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/browser-security-reviewer.agent.md

@@ -70,3 +70,23 @@ Return:
 - **Safe-to-merge conditions:** exact checks that must pass
 
 If there are no genuine issues, say so directly and do not invent style feedback.
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/dev-leader.agent.md

@@ -96,3 +96,23 @@ When you encounter a problem that spans domains:
 Primary: `*.py` (root tools), `hooks/**/*`, `browse/**/*.py`, `migrate.py`, `sk.py`, `install.py`
 
 Out of scope: `browse-ui/src/**/*` (belongs to browse-leader), `crates/**/*` (flag to orchestrator)
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/hosted-shell-bootstrap.agent.md

@@ -124,3 +124,23 @@ Open a PR with:
 - Test output
 - Browser smoke evidence if possible
 - Any follow-up issue needed for HTTPS companion or richer pairing
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/python-browse-backend.agent.md

@@ -103,3 +103,23 @@ If `watch-sessions.py` changes, also run or confirm coverage for `tests/test_wat
 ## Output
 
 Report the changed routes, security decisions, synchronization decisions, and test evidence. If any behavior is deferred, open or reference a follow-up issue instead of leaving silent TODOs.
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/qa-leader.agent.md

@@ -139,3 +139,23 @@ Security: no SQL interpolation, no pickle, no secrets
 Primary: All changed surfaces (read-only audit + gate execution)
 
 Out of scope: Implementing fixes (that belongs to dev-leader or browse-leader)
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/research-planner.agent.md

@@ -77,3 +77,23 @@ When creating an implementation issue, include:
 ## Output
 
 Produce concise research that is implementation-ready. If asked to create GitHub issues, make each issue specific enough that a cloud agent can implement it without hidden context.
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/test-leader.agent.md

@@ -113,3 +113,23 @@ Handoff must include: test suite output (pass/fail counts), coverage gaps identi
 Primary: `test_*.py`, `run_all_tests.py`, `tests/**/*`, `browse-ui/src/**/*.test.*`, `browse-ui/e2e/**/*`
 
 Review scope (read-only audit): all files changed by dev-leader or browse-leader
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/verification-gate.agent.md

@@ -90,3 +90,23 @@ Return:
 - **Sync coverage:** watcher, auto-update/install, docs, hooks, skills/agents, conventions, CI/deploy status
 - **Failures:** root cause and affected file/test when known
 - **Next action:** exact command or issue/PR comment to run next
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/agents/whole-app-impact-auditor.agent.md

@@ -91,3 +91,23 @@ Return:
 - **Verification matrix:** command or check per impacted surface
 - **Risks if skipped:** concrete breakage scenario
 - **Follow-up issues:** only when deferring work is safe and explicit
+
+## Harness Integration
+
+Quality over speed — always run harness gates before marking work done:
+
+- **If `harness.yaml` exists**: run `sk harness check` before writing handoff; verify all success criteria are green
+- **For new project setup tasks**: include `sk harness init` as part of setup to define success criteria upfront
+- **For complex agent dispatch tasks**: enable telemetry with `SK_HARNESS=1 sk <command>`
+- **Never skip gates to save time**: a 36% performance gap separates harnessed from un-harnessed work (CORE benchmark, arXiv 2412.04524)
+
+```bash
+# Check harness (when harness.yaml present)
+sk harness check
+
+# Enable middleware for agent task
+SK_HARNESS=1 sk briefing "my task"
+
+# Set up harness on new project
+sk harness init --yes
+```

.github/copilot-instructions.md

@@ -206,6 +206,22 @@ On Windows (PowerShell), apply these rules to reduce token consumption:
 
 **Closeout:** attach command output (not just assertions) · `sk learn` before `task_complete` · subagents handoff with `--status DONE --changed-file <file> --learn`.
 
+## Harness Engineering
+
+The 🛡️ 7 harness principles from `AGENTS.md` apply at runtime. Executable commands:
+
+| Principle | When | Command |
+|-----------|------|---------|
+| No-Ship-Bugs | Before every commit | `python3 test_security.py && python3 test_fixes.py` |
+| Follow-Workflow | New project setup | `sk harness init` |
+| Quality-Over-Speed | After implementation | `sk harness check` |
+| Tentacle-Orchestration | ≥3 files changed | `sk tentacle create <name> --briefing` |
+| No-Abandon | confidence < 1.0 | `sk briefing "<topic>"` → research loop |
+| Rules-First | Before every task | `sk briefing --auto --compact` |
+| Knowledge-Recording | After bug fix/pattern | `sk learn --mistake "Title" "Details"` |
+
+Enable harness dispatch middleware: `SK_HARNESS=1 sk <command>`
+
 ## Testing
 
 ```bash