[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6HC3BRY5G4G4C794956Z8GF":3,"skill-same:01M6HC3BRY5G4G4C794956Z8GF":111},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":49,"commitSha":50,"skillDirectory":51,"forksCount":52,"repositoryHtmlUrl":53,"repositoryHomepageUrl":54,"ownerHtmlUrl":55,"ownerBlogUrl":56,"ownerDescription":57,"updatedAt":58,"install":59,"resources":72},"01M6HC3BRY5G4G4C794956Z8GF","defenseclaw-add-connector",null,"Use when adding or changing a DefenseClaw connector for a new AI agent (a vendor command hook, in-agent plugin, policy bridge, LLM proxy or ACP route). Covers the hook contract and native hook runner, guardrails, enterprise managed mode, skills\u002FMCP\u002Fplugins, discovery, observability, CLI setup, TUI, installers, docs, tests and CI. Gives the architecture, a route decision tree, the ordered end-to-end checklist, per-layer references with the mistakes this codebase has already made, and a worked example (GitHub Copilot CLI).","cisco-ai-defense\u002Fdefenseclaw","cisco-ai-defense","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F220944709?v=4","01M6HC38HDCQK9FGE6GXJ4H2ZT","Cisco AI Defense","01M6HC38NKX5MMEWMQXCXGP498",912,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"Connector","add","add connector","defenseclaw add connector","github copilot cli","cli","references","route","architecture","ai agent","MCP","copilot","ci","Guardrails","decision","end","mode","bridge","runner","github copilot","2026-11-10 10:12:30",false,"---\nname: defenseclaw-add-connector\ndescription: Use when adding or changing a DefenseClaw connector for a new AI agent (a vendor command hook, in-agent plugin, policy bridge, LLM proxy or ACP route). Covers the hook contract and native hook runner, guardrails, enterprise managed mode, skills\u002FMCP\u002Fplugins, discovery, observability, CLI setup, TUI, installers, docs, tests and CI. Gives the architecture, a route decision tree, the ordered end-to-end checklist, per-layer references with the mistakes this codebase has already made, and a worked example (GitHub Copilot CLI).\n---\n\n# Add or change a DefenseClaw connector\n\nA connector is how DefenseClaw governs one AI agent. The agent calls DefenseClaw through a vendor hook, plugin, proxy or ACP stream. The gateway turns each event into a guardrail verdict and sends the verdict back in the agent's native format. The same connector name also drives setup, teardown, inventory, discovery, telemetry, the enterprise guardian, the TUI and the docs.\n\nA connector ID ends up in hundreds of places: the retired-connector fold in `e508ba54` touched 247 files, and the Amp connector (`cf576356`) touched 241. Most of those places are closed lists with a silent default, such as OpenClaw paths, `\"unknown\"` telemetry, a generic proxy message or \"not certified on Windows\". A missed entry usually fails without an error, which is why this skill exists.\n\nVerified against `fix\u002Fenterprise-hardening` at `fc9e12a2`. The references name files and symbols, not line numbers. Find each one with `git grep -n '\u003Csymbol>'`. Source code and tests are authoritative; if this skill disagrees with them, trust the code and update the skill.\n\nThe commit SHAs cited here come from that branch's history. If the branch reached `main` as a squash merge, `git show \u003Csha>` fails in a fresh clone. Find the change instead with `git log -S '\u003Csymbol>'` on a symbol the same entry names, or in the pull request's commit list.\n\n## Architecture\n\n```\n                         AI agent (CLI, IDE or desktop app)\n                                        |\n     +-------------------+--------------+------------------+------------------+\n     | command hook      | in-agent plugin| policy bridge    | LLM proxy        | ACP (stdio)\n     | (most connectors) | (opencode, amp)| (omnigent)       | (openclaw,       | (kiro; zed\n     |                   |                |                  |  zeptoclaw)      |  via catalog)\n     v                   v                v                  v                  v\n hooks\u002F\u003Cid>-hook.sh  hooks\u002F\u003Cid>-plugin.*  hooks\u002Fomnigent-    proxy.go           internal\u002Facp\n or defenseclaw-hook rendered into the    policy.py          \u002Fc\u002F\u003Cname>\u002F...      enterprise acp\n hook --connector id agent's plugin dir                      (model tool_calls) (text only)\n        \\__________________ POST \u002Fapi\u002Fv1\u002F\u003Cid>\u002Fhook __________\u002F                   |\n                    (scoped hook token, trace and identity headers)              |\n                                   |                                             |\n  gateway  hook_register.go -> handleUnifiedConnectorHook -> handleAgentHook     |\n           normalize (generic keys, then HookProfile.Decode \u002F DecodeToolArgs)    |\n           evaluateAgentHook -> inspectTrustedToolPolicyCtx                      |\n             (actionfacts + CEL rule packs, CodeGuard, Cisco AI Defense, judge)  |\n           mapHookActionForProfile \u002F MapVerdict -> Respond ------------------- verdict\n                                   |\n  response {action, raw_action, severity, would_block, ..., \u003CResponseFieldName>: vendor body}\n                                   |\n  hook runner (hookexec spec, shell script or plugin) -> vendor stdout + exit code\n\n  Side channels keyed by the same ID: audit + OTel, inventory (skills\u002FMCP\u002Fplugins),\n  AI discovery, doctor\u002Fstatus, hook_contract_lock.json, enterprise guardian.\n```\n\nLifecycle ownership:\n\n```\ndefenseclaw setup \u003Cx>   (Python)                      defenseclaw-gateway   (Go)\n  contract + platform check (hook_contracts.json)\n  writes config.yaml + \u003Cdata_dir>\u002Fpicked_connector\n  restarts the gateway  ----------------------------> Connector.Setup(SetupOpts)\n                                                        render hooks\u002F*, back up and patch the\n                                                        vendor config, seal hook_contract_lock.json\n  waits for the lock   \u003C----------------------------\ndefenseclaw uninstall   ----------------------------> connector teardown | verify --connector \u003Cx>\n\nenterprise standalone guardian (root \u002F LocalSystem)\n  enterprisepolicy.RouteFor(connector, goos):\n    machine_policy  Target writes the vendor's admin-only machine file\n    per_user        a worker running as each enrolled user calls Connector.Setup(ManagedEnterprise)\n    acp             `defenseclaw-gateway enterprise acp` mediates the agent\n    unsupported\n```\n\n## Choose the integration kind\n\n| Kind | Existing examples | Go type | Artifact | Native runner style | Contract gate |\n|---|---|---|---|---|---|\n| Command hook, shell script on Unix, native exe on Windows or managed | cursor, copilot, devin, hermes, openhands, antigravity | `hookOnlyConnector` (`hook_only.go`; Devin's constructor is in `devin.go`) | `hooks\u002F\u003Cid>-hook.sh` plus a vendor config patch | `styleHookEcho` \u002F `styleHookEchoDecision` | versioned |\n| Command hook with rich vendor config and machine policy | claudecode, codex | own type (`ClaudeCodeConnector`, `CodexConnector`) | script + settings\u002FTOML patch | `styleClaudeCode` \u002F `styleCodex` | versioned |\n| Command hook + ACP | kiro | `KiroConnector` | `.kiro\u002Fhooks\u002F*.json` + CLI agent file | `styleHookDecisionStderr` | not gated |\n| In-agent plugin | opencode (JS), amp (TS) | `hookOnlyConnector{pluginArtifact:true}`; Amp embeds it in `AMPConnector` | `hooks\u002Fopencode-plugin.js`, `hooks\u002Famp-plugin.ts` | `stylePluginBridge` (managed OpenCode), `styleActionStderr` (amp) | versioned |\n| In-process policy bridge | omnigent | `OmnigentConnector` | `hooks\u002Fomnigent-policy.py` | none | versioned |\n| LLM proxy | openclaw, zeptoclaw | own types | proxy routes | none | not gated |\n\n## Decision tree\n\n```\n1. Does the agent run an external command for each lifecycle event (a hooks config)?\n   yes -> command hook.\n     a. Does it have rich settings plus a vendor machine-policy file (like Claude\u002FCodex)?\n        yes -> dedicated connector type. no -> a hookOnlyConnector constructor.\n     b. Is the event name missing from stdin?      -> bind it at Setup: `--event` (copilot)\n                                                      or an extra argv (antigravity on Unix).\n     c. How does the vendor read the result?       -> pick the decisionStyle:\n        stdout JSON decides (HookEcho) | exit 2 + JSON (HookEchoDecision) |\n        exit 2 + stderr, stdout goes to the context (HookDecisionStderr) | exit 2 + reason (ClaudeCode).\n     d. Does the vendor lack an exit-status or fail-closed contract, so a failed or\n        timed-out hook always counts as allow?                  -> failOpenOnly (copilot, hermes).\n2. No command hooks, but it loads code plugins from a directory?  -> in-agent plugin (opencode\u002Famp).\n3. It exposes an in-process policy API?                        -> policy bridge (omnigent).\n4. It speaks ACP over stdio?                                   -> internal\u002Facp catalog entry (kiro, zed),\n                                                                  optionally plus native hooks.\n5. Only its LLM traffic can be intercepted?                    -> proxy connector (openclaw pattern).\n\nPer OS:\n   Windows: does the host run hooks through PowerShell, cmd.exe, bash, or an exec argv?\n            -> a branch in hookInvocationCommandFor. WSL-only -> PlatformUnsupported with a reason.\n   macOS:   is it an app bundle? -> macos_bundle_binaries \u002F unixAgentAppBundleBinaries.\nEnterprise (per OS):\n   vendor machine-wide hook source that standard users can't write?  -> machine_policy Target\n   otherwise per_user; can users rewrite or add hooks?  -> foreign-hook guard (connectorSources),\n                                                           or a documented threat-model residual.\n   no way to deliver hooks on that OS?                   -> acp or unsupported, with a refusal reason.\n```\n\n## Rules that prevent the known bugs\n\n1. **ID:** use lowercase `[a-z0-9]+` with no separators. `DEFENSECLAW_FAIL_MODE_\u003CID>` strips `-` and `_`, so `open-hands` and `openhands` would collide.\n2. **Closed lists:** give the connector an explicit arm in every closed list, even when the answer is \"none\" (`return nil`, `[]`, `unsupportedSurface(...)`). Default arms fall back to OpenClaw paths, `\"unknown\"`, a proxy message or \"not certified\".\n3. **Contract parity:** the Go contract (`builtinHookContracts`) and `cli\u002Fdefenseclaw\u002Finventory\u002Fhook_contracts.json` must agree on darwin, linux and windows. `HookScriptVersion` must equal the script's `# defenseclaw-managed-hook vN` marker (the plugin's `\u002F\u002F defenseclaw-managed-plugin vN`, the policy bridge's `# defenseclaw-managed-policy vN`). No test checks that they match, so check it yourself. Verify and doctor code also hard-codes some markers as literals, so grep for `defenseclaw-managed-` whenever you bump one ([core.md §12](reference\u002Fcore.md#12-known-gaps-check-before-relying-on-them)).\n4. **Event casing:** use the vendor's exact wire casing. Lifecycle routing, `foreignHookStopEvent` and `correlationLifecycleForContract` match exactly. OpenHands sent PascalCase `event_type` while the contract said snake_case, so PreToolUse was never inspected (`bb78ad66`).\n5. **Shell tools:** map every shell or exec tool name in all the shell tables, and project vendor metadata keys away so ActionFacts stays authoritative. If you don't, the result is a CRITICAL finding with `raw_action=allow`, not a block (Amp `async_shell_command` `d01406ee`, Kiro `execute_bash`, Antigravity `run_command` metadata).\n6. **Rendered arguments:** every argument in a rendered hook command must exist on the `defenseclaw-hook hook` command (`internal\u002Fcli\u002Fhook.go`). Kiro's `--hook-surface v3` was a usage error, exit 1, and Kiro went ahead (`cb96bc5d`). Never append positional arguments to the native runner; it is `cobra.NoArgs`.\n7. **Windows exit codes:** `defenseclaw-hook.exe` is a GUI-subsystem binary, so `& '\u003Cexe>'` does not wait and loses exit 2. Use the encoded `Process.Start` bridge (`windowsAwaitedHookStatements`) or a `.ps1` adapter, and test through the host's real shell (`cmd.exe \u002Fd \u002Fs \u002Fc`, PowerShell or bash) (`cd6b6f2c`).\n8. **Stop events:** put the vendor's stop and session-end events in `foreignHookStopEvent`. Otherwise a fail-closed standalone session loops on the stop hook.\n9. **Teardown:** teardown must recognize every command form ever rendered (legacy Windows forms, script versus admin binary). Keep old forms as teardown identities.\n10. **Enterprise guard:** a guarded connector must run the admin-owned `defenseclaw-hook ... --enterprise-managed` (or call `--foreign-hook-check`), never a user-writable script (Devin, `0f728f8b`).\n11. **Secure Client:** the Secure Client profile must stay byte-identical. Do not change frozen Secure Client assets without an explicitly requested scope. The requested OpenCode V1 removal ports its Secure Client plugin to the shared V2-only adapter; its account\u002Fservice verification remains unchanged. Other `*-secure-client.*` templates, the three-connector Windows certified set and the macOS allow-list stay frozen.\n12. **Hygiene:** don't hand-edit generated `openwiki\u002F` pages. Don't spell retired connector names; `cli\u002Ftests\u002Ftest_retired_connector_names.py` scans every tracked file. Keep host names, accounts and local paths out of the repository.\n\n## End-to-end order\n\nWork in this order. Each step links to the reference section with the file- and symbol-level detail. [reference\u002Fchecklist.md](reference\u002Fchecklist.md) is the copyable version, with one line per touch point and the test that verifies it.\n\n0. **Research the vendor** before you write any code. Record, with doc URLs and a pinned revision:\n   - the hook events and their exact casing;\n   - stdin shapes for every event, captured from a real run;\n   - how the agent reads the result (stdout, exit code, stderr);\n   - hook ordering and merge rules when several sources register hooks;\n   - config paths per OS and the environment variables that redirect them;\n   - the version command;\n   - where MCP servers, skills, plugins, rules and agents live;\n   - native OTel;\n   - any machine-policy source;\n   - shell and exec tool names and their argument keys.\n\n   These become the contract `Notes`, the correlation provenance and the tests.\n1. **Identity and the Go type.** Pick the ID, aliases and derived names. Implement `Connector` and the optional interfaces. Register it in `newBuiltinConnectors` and add a `windowsConnectorSupport` row. See [core.md §1–2](reference\u002Fcore.md#1-identity).\n2. **Hook contract.** Add the Go contract, the JSON manifest entry, the tool-call lifecycle, the correlation spec, and the per-OS pins. See [core.md §3](reference\u002Fcore.md#3-hook-contract-and-lifecycle).\n3. **Runtime artifact.** Add the script, plugin or adapter template; `connectorHookScripts`; the `hookexec` spec and dialect branches; the Windows command; and the owned-command needles. See [core.md §4–6](reference\u002Fcore.md#4-hook-templates).\n4. **Setup, teardown and verify.** Cover backups, idempotence, tombstones, the contract lock and fail mode. See [core.md §7](reference\u002Fcore.md#7-setup-teardown-verify).\n5. **Gateway and guardrails.** Add the route; Decode and DecodeToolArgs; event classifiers; shell-tool normalization; the MCP and skill identity; verdict mapping and the response. See [guardrails.md](reference\u002Fguardrails.md).\n6. **Assets.** Add capabilities and `ComponentTargets`; the Go `claw.go` resolvers; managed MCP inventory; Python `connector_paths.py`; the AIBOM; plugin manifests; self-exemption for bridges. See [assets.md](reference\u002Fassets.md).\n7. **Discovery and observability.** Add AI signatures (Go and Python copies byte-identical); agent discovery specs; endpoint inventory; schema enums; telemetry normalizers; correlation; native OTLP; audit identity; health and status names; dashboards. See [discovery-observability.md](reference\u002Fdiscovery-observability.md).\n8. **CLI, TUI and installers.** Add setup tables and branches; onboarding; doctor, status and fail mode; uninstall and migration; TUI pickers and labels; the Go CLI; env vars; `install.sh` and `install.ps1` lists; the macOS app's connector lists. See [cli-tui.md](reference\u002Fcli-tui.md).\n9. **Enterprise managed mode.** Add the route per OS; the machine-policy target or per-user enrollment; the guard; the version probe and floor; every hard-coded Windows per-user list; ACP; threat-model rows. See [enterprise.md](reference\u002Fenterprise.md).\n10. **Docs.** Add the connector page; `meta.json`; the index and compatibility rows; the capability matrix; the icon; the pages that list every connector; enterprise docs; CHANGELOG. See [docs-tests-ci.md §1–3](reference\u002Fdocs-tests-ci.md#1-docs).\n11. **Tests and CI.** Update the hard-coded lists in tests; the registry-driven parity tests; per-connector tests; the e2e matrix row and golden; the live E2E harness; CI matrices and the tests that pin them. See [docs-tests-ci.md §4–6](reference\u002Fdocs-tests-ci.md#4-tests).\n12. **Validate** with the narrowest commands that prove the change ([docs-tests-ci.md §7](reference\u002Fdocs-tests-ci.md#7-validation-commands)). Then certify live on each OS you claim, and only then add a `validated_versions.json` row.\n\n## Find every list: the sibling grep\n\nNo single registry drives everything. The reliable way to find every touch point is to pick the existing connector closest to yours, list every place it appears, and decide for each hit whether your connector belongs there:\n\n```bash\n# closest sibling: copilot (per-event command hook + machine policy), devin (per-user hook),\n# amp\u002Fopencode (plugin), kiro (ACP + hooks), omnigent (policy bridge), openclaw (proxy)\ngit grep -n -i -w '\u003Csibling>' -- ':!openwiki' ':!CHANGELOG.md' | wc -l\ngit grep -l -i -w '\u003Csibling>' -- ':!openwiki' | awk -F\u002F '{print $1\"\u002F\"$2}' | sort | uniq -c | sort -rn\n# then, per directory, review each hit\ngit grep -n -i -w '\u003Csibling>' -- internal\u002Fenterprisehooks\n```\n\nAlso grep for lists that hold several connectors but not your sibling, for example `git grep -n '\"hermes\", *\"opencode\"'`. The worked example, [reference\u002Fworked-example.md](reference\u002Fworked-example.md), is this search already done for Copilot.\n\n## Reference files\n\n| File | Covers |\n|---|---|\n| [reference\u002Fcore.md](reference\u002Fcore.md) | identity, Go interfaces, registry, hook contract, lifecycle, correlation, templates, `hookexec`, Windows commands, setup\u002Fteardown\u002Fverify, fail modes, exit codes |\n| [reference\u002Fguardrails.md](reference\u002Fguardrails.md) | request pipeline, payload decode, event classes, ActionFacts shell tables, proof gating, verdict mapping, vendor block\u002Fask rendering |\n| [reference\u002Fenterprise.md](reference\u002Fenterprise.md) | routes, machine-policy targets, foreign-hook guard, Unix guardian and worker, Windows standalone lists, floors, ACP, Secure Client invariants, threat model |\n| [reference\u002Fassets.md](reference\u002Fassets.md) | skills, MCP servers, plugins, rules and agents: capabilities, watcher, config resolvers, inventory, AIBOM, bridge self-exemption |\n| [reference\u002Fdiscovery-observability.md](reference\u002Fdiscovery-observability.md) | AI signatures, agent discovery, endpoint inventory, schemas, telemetry normalizers, correlation, native OTLP, audit, health, dashboards |\n| [reference\u002Fcli-tui.md](reference\u002Fcli-tui.md) | Python setup, onboarding, doctor, status, fail mode, uninstall, migration, TUI, Go CLI, env vars, installers |\n| [reference\u002Fdocs-tests-ci.md](reference\u002Fdocs-tests-ci.md) | docs pages and data, hard-coded test lists, registry-driven parity tests, e2e matrix, live E2E harness, CI workflows, validation commands |\n| [reference\u002Fworked-example.md](reference\u002Fworked-example.md) | GitHub Copilot CLI traced through every layer, file by file in order |\n| [reference\u002Fchecklist.md](reference\u002Fchecklist.md) | copyable checklist, one line per touch point, with the verifying test or command |\n\n## Study these commits\n\n- `cf576356`: Amp added as a policy plugin, end to end.\n- `f29b76ec`: OmniGent policy connector.\n- `f05a9d3c`: ACP guard with Kiro and Zed.\n- `e508ba54` (the port of #903): a connector folded into another. It shows every place a connector ID lives, including the config-migration maps.\n\nThe fix commits named in each reference show the mistakes to avoid.\n\nRun `git show --stat \u003Csha>` to see a commit's footprint. The first three are on `main`; the others may need the `git log -S` search described at the top.\n","# Add or change a DefenseClaw connector\n\nA connector is how DefenseClaw governs one AI agent. The agent calls DefenseClaw through a vendor hook, plugin, proxy or ACP stream. The gateway turns each event into a guardrail verdict and sends the verdict back in the agent's native format. The same connector name also drives setup, teardown, inventory, discovery, telemetry, the enterprise guardian, the TUI and the docs.\n\nA connector ID ends up in hundreds of places: the retired-connector fold in `e508ba54` touched 247 files, and the Amp connector (`cf576356`) touched 241. Most of those places are closed lists with a silent default, such as OpenClaw paths, `\"unknown\"` telemetry, a generic proxy message or \"not certified on Windows\". A missed entry usually fails without an error, which is why this skill exists.\n\nVerified against `fix\u002Fenterprise-hardening` at `fc9e12a2`. The references name files and symbols, not line numbers. Find each one with `git grep -n '\u003Csymbol>'`. Source code and tests are authoritative; if this skill disagrees with them, trust the code and update the skill.\n\nThe commit SHAs cited here come from that branch's history. If the branch reached `main` as a squash merge, `git show \u003Csha>` fails in a fresh clone. Find the change instead with `git log -S '\u003Csymbol>'` on a symbol the same entry names, or in the pull request's commit list.\n\n## Architecture\n\n```\n                         AI agent (CLI, IDE or desktop app)\n                                        |\n     +-------------------+--------------+------------------+------------------+\n     | command hook      | in-agent plugin| policy bridge    | LLM proxy        | ACP (stdio)\n     | (most connectors) | (opencode, amp)| (omnigent)       | (openclaw,       | (kiro; zed\n     |                   |                |                  |  zeptoclaw)      |  via catalog)\n     v                   v                v                  v                  v\n hooks\u002F\u003Cid>-hook.sh  hooks\u002F\u003Cid>-plugin.*  hooks\u002Fomnigent-    proxy.go           internal\u002Facp\n or defenseclaw-hook rendered into the    policy.py          \u002Fc\u002F\u003Cname>\u002F...      enterprise acp\n hook --connector id agent's plugin dir                      (model tool_calls) (text only)\n        \\__________________ POST \u002Fapi\u002Fv1\u002F\u003Cid>\u002Fhook __________\u002F                   |\n                    (scoped hook token, trace and identity headers)              |\n                                   |                                             |\n  gateway  hook_register.go -> handleUnifiedConnectorHook -> handleAgentHook     |\n           normalize (generic keys, then HookProfile.Decode \u002F DecodeToolArgs)    |\n           evaluateAgentHook -> inspectTrustedToolPolicyCtx                      |\n             (actionfacts + CEL rule packs, CodeGuard, Cisco AI Defense, judge)  |\n           mapHookActionForProfile \u002F MapVerdict -> Respond ------------------- verdict\n                                   |\n  response {action, raw_action, severity, would_block, ..., \u003CResponseFieldName>: vendor body}\n                                   |\n  hook runner (hookexec spec, shell script or plugin) -> vendor stdout + exit code\n\n  Side channels keyed by the same ID: audit + OTel, inventory (skills\u002FMCP\u002Fplugins),\n  AI discovery, doctor\u002Fstatus, hook_contract_lock.json, enterprise guardian.\n```\n\nLifecycle ownership:\n\n```\ndefenseclaw setup \u003Cx>   (Python)                      defenseclaw-gateway   (Go)\n  contract + platform check (hook_contracts.json)\n  writes config.yaml + \u003Cdata_dir>\u002Fpicked_connector\n  restarts the gateway  ----------------------------> Connector.Setup(SetupOpts)\n                                                        render hooks\u002F*, back up and patch the\n                                                        vendor config, seal hook_contract_lock.json\n  waits for the lock   \u003C----------------------------\ndefenseclaw uninstall   ----------------------------> connector teardown | verify --connector \u003Cx>\n\nenterprise standalone guardian (root \u002F LocalSystem)\n  enterprisepolicy.RouteFor(connector, goos):\n    machine_policy  Target writes the vendor's admin-only machine file\n    per_user        a worker running as each enrolled user calls Connector.Setup(ManagedEnterprise)\n    acp             `defenseclaw-gateway enterprise acp` mediates the agent\n    unsupported\n```\n\n## Choose the integration kind\n\n| Kind | Existing examples | Go type | Artifact | Native runner style | Contract gate |\n|---|---|---|---|---|---|\n| Command hook, shell script on Unix, native exe on Windows or managed | cursor, copilot, devin, hermes, openhands, antigravity | `hookOnlyConnector` (`hook_only.go`; Devin's constructor is in `devin.go`) | `hooks\u002F\u003Cid>-hook.sh` plus a vendor config patch | `styleHookEcho` \u002F `styleHookEchoDecision` | versioned |\n| Command hook with rich vendor config and machine policy | claudecode, codex | own type (`ClaudeCodeConnector`, `CodexConnector`) | script + settings\u002FTOML patch | `styleClaudeCode` \u002F `styleCodex` | versioned |\n| Command hook + ACP | kiro | `KiroConnector` | `.kiro\u002Fhooks\u002F*.json` + CLI agent file | `styleHookDecisionStderr` | not gated |\n| In-agent plugin | opencode (JS), amp (TS) | `hookOnlyConnector{pluginArtifact:true}`; Amp embeds it in `AMPConnector` | `hooks\u002Fopencode-plugin.js`, `hooks\u002Famp-plugin.ts` | `stylePluginBridge` (managed OpenCode), `styleActionStderr` (amp) | versioned |\n| In-process policy bridge | omnigent | `OmnigentConnector` | `hooks\u002Fomnigent-policy.py` | none | versioned |\n| LLM proxy | openclaw, zeptoclaw | own types | proxy routes | none | not gated |\n\n## Decision tree\n\n```\n1. Does the agent run an external command for each lifecycle event (a hooks config)?\n   yes -> command hook.\n     a. Does it have rich settings plus a vendor machine-policy file (like Claude\u002FCodex)?\n        yes -> dedicated connector type. no -> a hookOnlyConnector constructor.\n     b. Is the event name missing from stdin?      -> bind it at Setup: `--event` (copilot)\n                                                      or an extra argv (antigravity on Unix).\n     c. How does the vendor read the result?       -> pick the decisionStyle:\n        stdout JSON decides (HookEcho) | exit 2 + JSON (HookEchoDecision) |\n        exit 2 + stderr, stdout goes to the context (HookDecisionStderr) | exit 2 + reason (ClaudeCode).\n     d. Does the vendor lack an exit-status or fail-closed contract, so a failed or\n        timed-out hook always counts as allow?                  -> failOpenOnly (copilot, hermes).\n2. No command hooks, but it loads code plugins from a directory?  -> in-agent plugin (opencode\u002Famp).\n3. It exposes an in-process policy API?                        -> policy bridge (omnigent).\n4. It speaks ACP over stdio?                                   -> internal\u002Facp catalog entry (kiro, zed),\n                                                                  optionally plus native hooks.\n5. Only its LLM traffic can be intercepted?                    -> proxy connector (openclaw pattern).\n\nPer OS:\n   Windows: does the host run hooks through PowerShell, cmd.exe, bash, or an exec argv?\n            -> a branch in hookInvocationCommandFor. WSL-only -> PlatformUnsupported with a reason.\n   macOS:   is it an app bundle? -> macos_bundle_binaries \u002F unixAgentAppBundleBinaries.\nEnterprise (per OS):\n   vendor machine-wide hook source that standard users can't write?  -> machine_policy Target\n   otherwise per_user; can users rewrite or add hooks?  -> foreign-hook guard (connectorSources),\n                                                           or a documented threat-model residual.\n   no way to deliver hooks on that OS?                   -> acp or unsupported, with a refusal reason.\n```\n\n## Rules that prevent the known bugs\n\n1. **ID:** use lowercase `[a-z0-9]+` with no separators. `DEFENSECLAW_FAIL_MODE_\u003CID>` strips `-` and `_`, so `open-hands` and `openhands` would collide.\n2. **Closed lists:** give the connector an explicit arm in every closed list, even when the answer is \"none\" (`return nil`, `[]`, `unsupportedSurface(...)`). Default arms fall back to OpenClaw paths, `\"unknown\"`, a proxy message or \"not certified\".\n3. **Contract parity:** the Go contract (`builtinHookContracts`) and `cli\u002Fdefenseclaw\u002Finventory\u002Fhook_contracts.json` must agree on darwin, linux and windows. `HookScriptVersion` must equal the script's `# defenseclaw-managed-hook vN` marker (the plugin's `\u002F\u002F defenseclaw-managed-plugin vN`, the policy bridge's `# defenseclaw-managed-policy vN`). No test checks that they match, so check it yourself. Verify and doctor code also hard-codes some markers as literals, so grep for `defenseclaw-managed-` whenever you bump one ([core.md §12](reference\u002Fcore.md#12-known-gaps-check-before-relying-on-them)).\n4. **Event casing:** use the vendor's exact wire casing. Lifecycle routing, `foreignHookStopEvent` and `correlationLifecycleForContract` match exactly. OpenHands sent PascalCase `event_type` while the contract said snake_case, so PreToolUse was never inspected (`bb78ad66`).\n5. **Shell tools:** map every shell or exec tool name in all the shell tables, and project vendor metadata keys away so ActionFacts stays authoritative. If you don't, the result is a CRITICAL finding with `raw_action=allow`, not a block (Amp `async_shell_command` `d01406ee`, Kiro `execute_bash`, Antigravity `run_command` metadata).\n6. **Rendered arguments:** every argument in a rendered hook command must exist on the `defenseclaw-hook hook` command (`internal\u002Fcli\u002Fhook.go`). Kiro's `--hook-surface v3` was a usage error, exit 1, and Kiro went ahead (`cb96bc5d`). Never append positional arguments to the native runner; it is `cobra.NoArgs`.\n7. **Windows exit codes:** `defenseclaw-hook.exe` is a GUI-subsystem binary, so `& '\u003Cexe>'` does not wait and loses exit 2. Use the encoded `Process.Start` bridge (`windowsAwaitedHookStatements`) or a `.ps1` adapter, and test through the host's real shell (`cmd.exe \u002Fd \u002Fs \u002Fc`, PowerShell or bash) (`cd6b6f2c`).\n8. **Stop events:** put the vendor's stop and session-end events in `foreignHookStopEvent`. Otherwise a fail-closed standalone session loops on the stop hook.\n9. **Teardown:** teardown must recognize every command form ever rendered (legacy Windows forms, script versus admin binary). Keep old forms as teardown identities.\n10. **Enterprise guard:** a guarded connector must run the admin-owned `defenseclaw-hook ... --enterprise-managed` (or call `--foreign-hook-check`), never a user-writable script (Devin, `0f728f8b`).\n11. **Secure Client:** the Secure Client profile must stay byte-identical. Do not change frozen Secure Client assets without an explicitly requested scope. The requested OpenCode V1 removal ports its Secure Client plugin to the shared V2-only adapter; its account\u002Fservice verification remains unchanged. Other `*-secure-client.*` templates, the three-connector Windows certified set and the macOS allow-list stay frozen.\n12. **Hygiene:** don't hand-edit generated `openwiki\u002F` pages. Don't spell retired connector names; `cli\u002Ftests\u002Ftest_retired_connector_names.py` scans every tracked file. Keep host names, accounts and local paths out of the repository.\n\n## End-to-end order\n\nWork in this order. Each step links to the reference section with the file- and symbol-level detail. [reference\u002Fchecklist.md](reference\u002Fchecklist.md) is the copyable version, with one line per touch point and the test that verifies it.\n\n0. **Research the vendor** before you write any code. Record, with doc URLs and a pinned revision:\n   - the hook events and their exact casing;\n   - stdin shapes for every event, captured from a real run;\n   - how the agent reads the result (stdout, exit code, stderr);\n   - hook ordering and merge rules when several sources register hooks;\n   - config paths per OS and the environment variables that redirect them;\n   - the version command;\n   - where MCP servers, skills, plugins, rules and agents live;\n   - native OTel;\n   - any machine-policy source;\n   - shell and exec tool names and their argument keys.\n\n   These become the contract `Notes`, the correlation provenance and the tests.\n1. **Identity and the Go type.** Pick the ID, aliases and derived names. Implement `Connector` and the optional interfaces. Register it in `newBuiltinConnectors` and add a `windowsConnectorSupport` row. See [core.md §1–2](reference\u002Fcore.md#1-identity).\n2. **Hook contract.** Add the Go contract, the JSON manifest entry, the tool-call lifecycle, the correlation spec, and the per-OS pins. See [core.md §3](reference\u002Fcore.md#3-hook-contract-and-lifecycle).\n3. **Runtime artifact.** Add the script, plugin or adapter template; `connectorHookScripts`; the `hookexec` spec and dialect branches; the Windows command; and the owned-command needles. See [core.md §4–6](reference\u002Fcore.md#4-hook-templates).\n4. **Setup, teardown and verify.** Cover backups, idempotence, tombstones, the contract lock and fail mode. See [core.md §7](reference\u002Fcore.md#7-setup-teardown-verify).\n5. **Gateway and guardrails.** Add the route; Decode and DecodeToolArgs; event classifiers; shell-tool normalization; the MCP and skill identity; verdict mapping and the response. See [guardrails.md](reference\u002Fguardrails.md).\n6. **Assets.** Add capabilities and `ComponentTargets`; the Go `claw.go` resolvers; managed MCP inventory; Python `connector_paths.py`; the AIBOM; plugin manifests; self-exemption for bridges. See [assets.md](reference\u002Fassets.md).\n7. **Discovery and observability.** Add AI signatures (Go and Python copies byte-identical); agent discovery specs; endpoint inventory; schema enums; telemetry normalizers; correlation; native OTLP; audit identity; health and status names; dashboards. See [discovery-observability.md](reference\u002Fdiscovery-observability.md).\n8. **CLI, TUI and installers.** Add setup tables and branches; onboarding; doctor, status and fail mode; uninstall and migration; TUI pickers and labels; the Go CLI; env vars; `install.sh` and `install.ps1` lists; the macOS app's connector lists. See [cli-tui.md](reference\u002Fcli-tui.md).\n9. **Enterprise managed mode.** Add the route per OS; the machine-policy target or per-user enrollment; the guard; the version probe and floor; every hard-coded Windows per-user list; ACP; threat-model rows. See [enterprise.md](reference\u002Fenterprise.md).\n10. **Docs.** Add the connector page; `meta.json`; the index and compatibility rows; the capability matrix; the icon; the pages that list every connector; enterprise docs; CHANGELOG. See [docs-tests-ci.md §1–3](reference\u002Fdocs-tests-ci.md#1-docs).\n11. **Tests and CI.** Update the hard-coded lists in tests; the registry-driven parity tests; per-connector tests; the e2e matrix row and golden; the live E2E harness; CI matrices and the tests that pin them. See [docs-tests-ci.md §4–6](reference\u002Fdocs-tests-ci.md#4-tests).\n12. **Validate** with the narrowest commands that prove the change ([docs-tests-ci.md §7](reference\u002Fdocs-tests-ci.md#7-validation-commands)). Then certify live on each OS you claim, and only then add a `validated_versions.json` row.\n\n## Find every list: the sibling grep\n\nNo single registry drives everything. The reliable way to find every touch point is to pick the existing connector closest to yours, list every place it appears, and decide for each hit whether your connector belongs there:\n\n```bash\n# closest sibling: copilot (per-event command hook + machine policy), devin (per-user hook),\n# amp\u002Fopencode (plugin), kiro (ACP + hooks), omnigent (policy bridge), openclaw (proxy)\ngit grep -n -i -w '\u003Csibling>' -- ':!openwiki' ':!CHANGELOG.md' | wc -l\ngit grep -l -i -w '\u003Csibling>' -- ':!openwiki' | awk -F\u002F '{print $1\"\u002F\"$2}' | sort | uniq -c | sort -rn\n# then, per directory, review each hit\ngit grep -n -i -w '\u003Csibling>' -- internal\u002Fenterprisehooks\n```\n\nAlso grep for lists that hold several connectors but not your sibling, for example `git grep -n '\"hermes\", *\"opencode\"'`. The worked example, [reference\u002Fworked-example.md](reference\u002Fworked-example.md), is this search already done for Copilot.\n\n## Reference files\n\n| File | Covers |\n|---|---|\n| [reference\u002Fcore.md](reference\u002Fcore.md) | identity, Go interfaces, registry, hook contract, lifecycle, correlation, templates, `hookexec`, Windows commands, setup\u002Fteardown\u002Fverify, fail modes, exit codes |\n| [reference\u002Fguardrails.md](reference\u002Fguardrails.md) | request pipeline, payload decode, event classes, ActionFacts shell tables, proof gating, verdict mapping, vendor block\u002Fask rendering |\n| [reference\u002Fenterprise.md](reference\u002Fenterprise.md) | routes, machine-policy targets, foreign-hook guard, Unix guardian and worker, Windows standalone lists, floors, ACP, Secure Client invariants, threat model |\n| [reference\u002Fassets.md](reference\u002Fassets.md) | skills, MCP servers, plugins, rules and agents: capabilities, watcher, config resolvers, inventory, AIBOM, bridge self-exemption |\n| [reference\u002Fdiscovery-observability.md](reference\u002Fdiscovery-observability.md) | AI signatures, agent discovery, endpoint inventory, schemas, telemetry normalizers, correlation, native OTLP, audit, health, dashboards |\n| [reference\u002Fcli-tui.md](reference\u002Fcli-tui.md) | Python setup, onboarding, doctor, status, fail mode, uninstall, migration, TUI, Go CLI, env vars, installers |\n| [reference\u002Fdocs-tests-ci.md](reference\u002Fdocs-tests-ci.md) | docs pages and data, hard-coded test lists, registry-driven parity tests, e2e matrix, live E2E harness, CI workflows, validation commands |\n| [reference\u002Fworked-example.md](reference\u002Fworked-example.md) | GitHub Copilot CLI traced through every layer, file by file in order |\n| [reference\u002Fchecklist.md](reference\u002Fchecklist.md) | copyable checklist, one line per touch point, with the verifying test or command |\n\n## Study these commits\n\n- `cf576356`: Amp added as a policy plugin, end to end.\n- `f29b76ec`: OmniGent policy connector.\n- `f05a9d3c`: ACP guard with Kiro and Zed.\n- `e508ba54` (the port of #903): a connector folded into another. It shows every place a connector ID lives, including the config-migration maps.\n\nThe fix commits named in each reference show the mistakes to avoid.\n\nRun `git show --stat \u003Csha>` to see a commit's footprint. The first three are on `main`; the others may need the `git log -S` search described at the top.\n",{"name":5,"description":7},"main","00e6d52cf34e7f32390000c712f11a91a2029ce0",".claude\u002Fskills\u002Fdefenseclaw-add-connector",166,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw","https:\u002F\u002Fcisco-ai-defense.github.io\u002Fdefenseclaw\u002F","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense","https:\u002F\u002Fcisco-ai-defense.github.io\u002F","Open Source projects for the Cisco AI Defense community","2026-11-15 03:53:21",{"source":8,"skillName":5,"commands":60,"prompt":70,"zipAvailable":71},[61,64,67],{"tool":62,"command":63},"npx","npx skills add cisco-ai-defense\u002Fdefenseclaw --skill defenseclaw-add-connector",{"tool":65,"command":66},"bunx","bunx skills add cisco-ai-defense\u002Fdefenseclaw --skill defenseclaw-add-connector",{"tool":68,"command":69},"pnpm","pnpm dlx skills add cisco-ai-defense\u002Fdefenseclaw --skill defenseclaw-add-connector","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002FSKILL.md",true,[73,79,83,87,91,95,99,103,107],{"path":74,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":77,"sourceHtmlUrl":78},"reference\u002Fassets.md","REFERENCE","参考文档",20144,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fassets.md",{"path":80,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":81,"sourceHtmlUrl":82},"reference\u002Fchecklist.md",19430,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fchecklist.md",{"path":84,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":85,"sourceHtmlUrl":86},"reference\u002Fcli-tui.md",18133,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fcli-tui.md",{"path":88,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":89,"sourceHtmlUrl":90},"reference\u002Fcore.md",34169,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fcore.md",{"path":92,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":93,"sourceHtmlUrl":94},"reference\u002Fdiscovery-observability.md",17185,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fdiscovery-observability.md",{"path":96,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":97,"sourceHtmlUrl":98},"reference\u002Fdocs-tests-ci.md",18579,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fdocs-tests-ci.md",{"path":100,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":101,"sourceHtmlUrl":102},"reference\u002Fenterprise.md",28253,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fenterprise.md",{"path":104,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":105,"sourceHtmlUrl":106},"reference\u002Fguardrails.md",22461,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fguardrails.md",{"path":108,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":109,"sourceHtmlUrl":110},"reference\u002Fworked-example.md",16577,"https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-add-connector\u002Freference\u002Fworked-example.md",[112,128,137,146,172,186],{"skillNo":113,"name":114,"nameZh":6,"description":115,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":116,"categories":117,"keywords":118,"publishedAt":44},"01M6HC44X1K8NMF6YBHDZHS9PB","parity-fixture","Fixture for the rule-pack parity test of install admission and skill scan.","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002Ftestdata\u002Frulepack_artifact_parity\u002Fskill\u002FSKILL.md",[],[119,120,121,122,123,124,125,126,127],"parity fixture","fixture","pack","for","test","skill scan","rule","scan","install",{"skillNo":129,"name":130,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":131,"categories":132,"keywords":134,"publishedAt":44},"01M6HC43H8K4P02PSEJ6R1DCAE","malicious-skill","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002Ftest\u002Ffixtures\u002Fskills\u002Fmalicious-skill\u002FSKILL.md",[133],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[135,136],"malicious","malicious skill",{"skillNo":138,"name":139,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":140,"categories":141,"keywords":143,"publishedAt":44},"01M6HC425BT2XS2J16M5PBE6YX","clean-skill","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002Ftest\u002Ffixtures\u002Fskills\u002Fclean-skill\u002FSKILL.md",[142],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[144,145],"clean","clean skill",{"skillNo":147,"name":148,"nameZh":6,"description":149,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":150,"categories":151,"keywords":157,"publishedAt":44},"01M6HC3W7VSGDPE5WR47ZNH04M","codeguard","Security-aware code generation — teaches the agent CodeGuard rules to write secure code by default","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002Fskills\u002Fcodeguard\u002FSKILL.md",[152],{"categoryNo":153,"categoryName":154,"taxonomyType":155,"taxonomyTypeDesc":156},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[148,158,159,160,161,162,163,164,49,165,166,167,168,169,170,171],"Code Generation","network","Rules","check","secure","write","default","self","local","generation","calls","gateway","security","code",{"skillNo":173,"name":174,"nameZh":6,"description":175,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":176,"categories":177,"keywords":178,"publishedAt":44},"01M6HC3SGQQZQZHNBF80QNTZNE","perf-hello","Prints a friendly greeting when the user asks to say hello.","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002Fscripts\u002Fperf\u002Ffixtures\u002Fbenign-skill\u002FSKILL.md",[],[179,180,181,182,183,184,185],"perf hello","perf","hello","say hello","user","say","greeting",{"skillNo":187,"name":188,"nameZh":6,"description":189,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":190,"categories":191,"keywords":195,"publishedAt":44},"01M6HC3KVG7M9ENR2EGZSXF05T","defenseclaw-tui","Improve the DefenseClaw Textual TUI (cli\u002Fdefenseclaw\u002Ftui) consistently. Use it for any TUI work - adding or changing panels, picker and consequence modals, Setup wizards and the config editor, the hint bar and help overlay, command-palette rows, TUI tests, and TUI UX reviews or render checks. It holds the golden rules, the file map, recipes, the 80x24 layout budget, the test budget, and scripts that render the screen headlessly, run a fake-data demo, audit CLI coverage and list a panel's touchpoints. Don't use it for live certification on the Windows\u002FmacOS\u002FLinux test hosts (use defenseclaw-live-testing) or for EC2 dev-host operations (use defenseclaw-dev-host).","https:\u002F\u002Fgithub.com\u002Fcisco-ai-defense\u002Fdefenseclaw\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdefenseclaw-tui\u002FSKILL.md",[192],{"categoryNo":193,"categoryName":194,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000008","测试与质量",[196,197,198,29,199,200,201,202,203,204,205,206,207,208,209,210,211,212,213,214],"defenseclaw tui","tui","consequence","run","ux","audit","improve","work","command","coverage","map","layout","host","panels","palette","render","reviews","hosts","testing"]