[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M754FHG3W44SYK2BBXPT5BDW":3,"skill-same:01M754FHG3W44SYK2BBXPT5BDW":54},{"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":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":32,"updatedAt":33,"install":34,"resources":47},"01M754FHG3W44SYK2BBXPT5BDW","mecatl-learning-config",null,"Configure, write, or tune the mecatl learning: settings policy, including learning.mode, sensitivity, reflection budgets, and validated\u002Fevaluated activation. NOT for running \u002Freflect, reviewing proposals or memories, drafting skills, model routing, credentials, evaluator implementation, or other harnesses.","stacklok\u002Fmecatl","stacklok","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F110237746?v=4","01M754F1F8WJJZYJ0WZ73C7PBS","Stacklok","01M754F1GGDR6FG8K1RZ8SR2NA",309,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fmecatl-learning-config\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: mecatl-learning-config\ndescription: >-\n  Configure, write, or tune the mecatl learning: settings policy, including\n  learning.mode, sensitivity, reflection budgets, and validated\u002Fevaluated\n  activation. NOT for running \u002Freflect, reviewing proposals or memories,\n  drafting skills, model routing, credentials, evaluator implementation, or\n  other harnesses.\n---\n\n# mecatl learning configuration\n\nSafely design and merge the operator-tier `learning:` policy after identifying\nwhere mecatl actually runs. Do not assume that the client and server share a\nhost or configuration.\n\n## Progressive reference use\n\nAfter deployment and target resolution (Workflow Step 1), read only the needed\nparts of [the configuration reference](references\u002Fconfig-format.md):\n\n- [Exact schema and defaults](references\u002Fconfig-format.md#exact-schema-and-defaults)\n  always;\n- [Budget profiles](references\u002Fconfig-format.md#budget-profiles) only after the\n  budget choice;\n- [Operator and project tiers](references\u002Fconfig-format.md#operator-and-project-tiers)\n  only when project configuration is discussed; and\n- [Common errors](references\u002Fconfig-format.md#common-errors) only when validating\n  or troubleshooting.\n\nDo not require reading the entire reference at activation. Repository background\nis in `user-docs\u002Freference\u002Fconfiguration.md`,\n`user-docs\u002Foperating\u002Fsettings.md`, and\n`user-docs\u002Ffeatures\u002Fagent-behavior\u002Flearning.md`; cite those paths as prose, not links.\n\n## Safety contract\n\n- Resolve the deployment, startup configuration, precedence, and effective target\n  before using Read on any settings file. Read only startup artifacts the operator\n  identifies or that are already available in the task context; never scan\n  processes\u002Fservices or unrelated files to discover them.\n- Read-only inspection and validator preflight are allowed before confirmation.\n  The proposed preflight may write only the generated, non-secret `learning:`\n  patch under the repository-local `.scratch\u002F`; it is not a settings write and\n  must be removed after validation on success, failure, or cancellation. Any\n  settings-file creation or modification requires explicit confirmation.\n- Inspect an established relevant target with the **Read tool**, never `cat`. Read\n  the complete file.\n  A missing file is fine. Do not print, log, or rewrite unrelated values,\n  comments, credentials, or secret-shaped content.\n- Treat malformed or unvalidated YAML as a stop condition. Do not edit until a\n  minimal repair is understood and explicitly approved.\n- Modify only the top-level `learning:` mapping. Preserve every other byte when\n  targeted replacement permits it. Never rewrite the whole file merely to make\n  YAML easier to generate.\n- Before any write, show the complete proposed `learning:` block and exact\n  replacement diff, then ask for explicit confirmation. Silence is not consent.\n- If the file is missing, offer manual content or creation of the resolved,\n  owner-intended path. Create it only after explicit confirmation. Never add\n  credentials.\n- The validator is read-only, uses no credentials or network, and must never\n  print file contents.\n\n## Workflow\n\n### 1. Establish deployment, startup configuration, and effective owner\n\nDo this before choosing or reading a settings path. First classify the deployment\nas local standalone, local embedded mecatui, remote client\u002Fserver, or engine\nembedder. Establish the actual startup source already supplied by the operator or\ntask: service command, container args\u002Fconfig, systemd unit\u002Foverrides, or embedder\ncode. From that source record, in order:\n\n1. every repeatable `--permission-config PATH` explicit operator file;\n2. whether `--permissions-conventional` is enabled; and\n3. the server process's `XDG_CONFIG_HOME`, or its `HOME` fallback when relevant.\n\nDo not infer these from defaults when the effective invocation may override them.\nIf the startup command\u002Funit\u002Fcontainer args are not observable from authorized,\nrelevant context, ask the operator for them. Never scan running processes,\nservices, containers, or unrelated files.\n\nResolve the owning operator learning block using mecatl's actual precedence:\n\n- inspect explicit files in CLI order, stopping at the first readable, valid file\n  with a non-null top-level `learning:` block; unreadable or invalid explicit files\n  are skipped by mecatl and cannot own the effective block;\n- only when conventional discovery is enabled and no explicit file captured a\n  learning block, consider the user file at\n  `$XDG_CONFIG_HOME\u002Fmecatl\u002Fsettings.yaml`, falling back to\n  `$HOME\u002F.config\u002Fmecatl\u002Fsettings.yaml` when `XDG_CONFIG_HOME` is unset or empty;\n  its readable, valid `learning:` block is then the owner; and\n- files after the owner, and a conventional user file shadowed by an explicit\n  owner, have no effect on operator learning. Never edit one of them.\n\nApply that resolution by deployment:\n\n- **Local standalone or embedded mecatui:** resolve against the local server's\n  actual invocation and environment, not merely the interactive shell's.\n- **Remote mecatui\u002Fconnect:** resolve only from server startup args and the server\n  host's environment\u002Ffiles. Never inspect or edit local client settings. Without\n  authorized server access, provide the exact block and server-side instructions.\n- **Engine embedder:** configuration is code-owned unless the embedding\n  application explicitly constructs `permconfig` from files. Ask the owner how\n  it is supplied; do not invent a `mecated` path.\n\nIf no current block owns learning, the insertion target is the first\noperator-intended explicit file chosen by the operator, or the conventional user\nfile only when conventional discovery is enabled. If ownership or an effective\ninsertion target cannot be established, do not edit automatically: provide a\nmanual block plus instructions to install and validate it on the owning server or\nin embedder code. Record the established path privately as `\u003Cresolved-path>`.\n\n### 2. Inspect and validate the current document\n\nRead the complete target and classify it as missing, valid without `learning:`,\nvalid with `learning:`, or malformed. Retain the existing mapping, comments,\nindentation, and byte range for a later targeted edit. Report only\nlearning-related findings.\n\nOn the server host, validate the target without exposing its content. Pass the\nresolved path as one quoted argument; never concatenate it into shell syntax, a\ncommand string, or another argument:\n\n```text\nmecated config validate --file \"\u003Cresolved-path>\"\n```\n\nThe command reads at most 256 KiB and prints only `valid` or a sanitized error.\nIt requires the file to exist unless a learning patch is supplied. If `mecated`\nis unavailable on a remote host, do not auto-edit: give the exact block and\nserver-side installation guidance, including actual-file validation after the\noperator makes the change.\n\n### 3. Elicit one answer at a time\n\nAsk in this order. Wait after every question. Each prompt must show all prior\nanswers and the recommended default:\n\n```text\n✓ Autonomy: \u003Canswer>\n✓ Sensitivity: \u003Canswer>\n\n→ \u003Cone current question>\n  ▸ \u003Crecommended choice> (recommended — \u003Cbrief reason>)\n    \u003Cother choices>\n```\n\n1. **Desired autonomy:** Review first \u002F Auto \u002F Off. Recommend **Review first**\n   for rollout. If the operator explicitly asks for autonomous learning,\n   recommend **Auto** instead. Off disables observation, not explicit tools.\n2. **Sensitivity:** conservative \u002F balanced \u002F eager. Recommend **balanced**.\n   This is independent of budget posture: any sensitivity may be combined with\n   any budget profile.\n3. **Skill activation assurance:** ask this when mode is Auto. Recommend\n   **validated** for usable, body-only, evidence-backed autonomy. Offer\n   **evaluated** only when a trusted fixture evaluator is configured; it\n   requires PASS. For Review\u002FOff, record future policy or accept validated as an\n   inert default without implying activation.\n4. **Budget posture:** default \u002F budget-conscious \u002F eager \u002F custom. Profiles set\n   only the six automatic budget values; they never choose sensitivity. Explain\n   that limits and cooldowns are process-local, so N replicas may reserve about\n   N times the aggregate. They are not provider quotas or cluster-global limits.\n   For custom, ask one automatic value at a time and validate it; do not ask for\n   sensitivity again.\n5. **Existing subtree:** when present, preserve or revise it. Recommend\n   **preserve and revise only selected fields**. When absent, state that a new\n   subtree will be inserted.\n\nSupport `back` to revise the preceding answer. Do not collapse these into one\nquestionnaire.\n\n### 4. Explain the selected behavior\n\nBefore proposing YAML, summarize mode × activation accurately:\n\n- **Off:** no automatic observation\u002Freflection. Explicit `\u002Freflect`, memory and\n  user-model tools, and `SkillDraft` remain available. Direct `SkillDraft`\n  creates inactive content. Dream consolidation is independently configured.\n- **Review:** admitted completions spend reflection capacity and stage durable\n  proposals; they do not promote memory or activate learned skills.\n- **Auto:** facts retain evidence, ownership, and conflict gates. With\n  `validated`, a structurally safe, body-only, evidence-backed exact candidate\n  activates on PASS or ABSTAIN\u002Fno evaluator. With `evaluated`, only trusted\n  evaluator PASS activates. FAIL and evaluator ERROR never activate.\n\nState separately that sensitivity selects automatic admission thresholds while\nbudgets bound automatic frequency and reserved tokens; users may combine any\nsensitivity with any budget posture. Explicit `\u002Freflect` bypasses automatic\nsensitivity, cooldown, and count\u002Ftoken budgets, but retains coordinator,\nprovider, timeout, ownership, and lifecycle limits.\n\n### 5. Preflight, propose, confirm, then edit\n\nGenerate a complete `learning:` block containing mode, independently selected\nsensitivity, skill activation, and all six automatic controls. Do not emit a\npartial subtree whose behavior depends on hidden defaults.\n\nBefore asking for confirmation, write only that exact non-secret `learning:`\nblock to a bounded, repo-local scratch name such as\n`.scratch\u002Flearning-preflight.yaml`, then run the read-only in-memory preflight\nwith both paths quoted:\n\n```text\nmecated config validate --file \"\u003Cresolved-path>\" \\\n  --learning-patch \".scratch\u002Flearning-preflight.yaml\"\n```\n\nThe scratch file is not the settings write: it contains only the exact block\nalready shown, never a full settings copy, credentials, or unrelated values. The\ncommand bounded-reads both files, requires the patch to be a single YAML document\nwith exactly one top-level `learning:` mapping, replaces or inserts only that node\nin memory, and validates the resulting complete document through mecatl's parser.\nIt prints `valid` or `valid (new file)` and never writes either input. Remove the\nscratch patch immediately after validation on success, failure, or cancellation.\nStop if preflight fails.\n\nIf the user requires no scratch write, provide the complete block for manual\napplication and require actual-file validation after the write instead; do not\nclaim a write-free proposed preflight was run.\n\nThen show:\n\n1. the complete `learning:` block;\n2. an exact diff replacing only the existing mapping, or exact insertion point;\n3. a concise effects\u002Fcost summary; and\n4. `Apply this exact change to \u003Cresolved-path>? (yes\u002Fno)`.\n\nOnly explicit yes authorizes an edit. Re-read the complete target immediately\nbefore editing. If it differs from the preflight input, stop, regenerate the\nlearning-only scratch patch, rerun `mecated config validate --file\n\"\u003Cresolved-path>\" --learning-patch \".scratch\u002F\u003Cbounded-name>.yaml\"` against the new\nbytes, remove the scratch patch, and ask again. Use a targeted exact replacement\npreserving unrelated YAML and comments. If safe targeting is impossible, offer\nmanual merge rather than rewriting the file.\n\nAfter writing, validate the actual file:\n\n```text\nmecated config validate --file \"\u003Cresolved-path>\"\n```\n\nA nonzero result is a failed application requiring immediate, minimal repair\nguidance; do not report success. Re-read and verify the selected learning values\nwithout showing unrelated content.\n\n### 6. Restart and observe\n\nLearning settings are resolved at build time: restart the local server, remote\nserver, or embedder instance that owns the resolved policy.\n\n- Local embedded mecatui: `\u002Flearning` cycles mode and\n  `\u002Flearning-sensitivity` cycles sensitivity; both save locally and still\n  require restart. Use the full file for budgets and activation assurance.\n- Remote mecatui\u002Fconnect: change and restart the server host, never the client.\n- Engine embedders: follow the application's configuration\u002Frebuild lifecycle.\n- After restart, inspect `\u002Freflections`, `\u002Fskills`, and `\u002Fusermodel`; use\n  `\u002Freflect` only when an intentional live run is desired. Warn that reflection\n  may make provider calls and incur token cost; verification need not force it.\n\nOffer rollback blocks from the reference: `mode: off`, `mode: review`, and the\nhigh-assurance `skills.activation: evaluated` tightening.\n\n## Error handling\n\n| Situation | Required response |\n|---|---|\n| Startup configuration not observable or ownership unresolved | Ask for the server command\u002Funit\u002Fcontainer args; otherwise supply a manual block and server\u002Fembedder validation instructions with no automatic edit. |\n| Earlier explicit file has a valid `learning:` block | It owns learning; do not edit a later explicit or conventional file. Explicit files are evaluated in CLI order. |\n| Explicit files have no captured block and conventional discovery is disabled | The XDG user file does not participate. Ask the operator to choose an explicit insertion target or provide a manual block. |\n| Missing file | Offer manual block or confirmed creation at the established `\u003Cresolved-path>`; do not guess ownership. |\n| Malformed, duplicate, unknown, or unvalidated YAML | Stop before editing; validate the complete document and obtain approval for minimal repair. |\n| Invalid value or duration | Reject using the exact schema; `window` must be `1m..24h`. |\n| Any maximum is zero | Warn that this bound disables all automatic reflection; cooldown zero only removes cooldown. |\n| Evaluated without evaluator | Warn that ABSTAIN\u002Fno-evaluator remains staged; recommend validated or separately configure a trusted evaluator. |\n| Project raises mode\u002Fsensitivity or loosens evaluated | Explain project policy is tighten-only and cannot raise operator autonomy. |\n| Project specifies automatic budgets | Explain budgets are operator-only and the project block is warning-ignored. |\n| Multiple replicas | Multiply the process-local envelope by replica count; never call it a cluster\u002Fprovider quota. |\n| Remote server inaccessible or `mecated` unavailable there | Supply the exact manual block and server-side instructions only; never inspect local client settings or auto-edit. Require `mecated config validate --file \"\u003Cresolved-path>\"` after installation when the binary becomes available. |\n| Routing, credentials, evaluator implementation, `\u002Freflect` execution, proposal\u002Fmemory review, skill drafting, or another harness | Decline that portion and route to its workflow. |\n","# mecatl learning configuration\n\nSafely design and merge the operator-tier `learning:` policy after identifying\nwhere mecatl actually runs. Do not assume that the client and server share a\nhost or configuration.\n\n## Progressive reference use\n\nAfter deployment and target resolution (Workflow Step 1), read only the needed\nparts of [the configuration reference](references\u002Fconfig-format.md):\n\n- [Exact schema and defaults](references\u002Fconfig-format.md#exact-schema-and-defaults)\n  always;\n- [Budget profiles](references\u002Fconfig-format.md#budget-profiles) only after the\n  budget choice;\n- [Operator and project tiers](references\u002Fconfig-format.md#operator-and-project-tiers)\n  only when project configuration is discussed; and\n- [Common errors](references\u002Fconfig-format.md#common-errors) only when validating\n  or troubleshooting.\n\nDo not require reading the entire reference at activation. Repository background\nis in `user-docs\u002Freference\u002Fconfiguration.md`,\n`user-docs\u002Foperating\u002Fsettings.md`, and\n`user-docs\u002Ffeatures\u002Fagent-behavior\u002Flearning.md`; cite those paths as prose, not links.\n\n## Safety contract\n\n- Resolve the deployment, startup configuration, precedence, and effective target\n  before using Read on any settings file. Read only startup artifacts the operator\n  identifies or that are already available in the task context; never scan\n  processes\u002Fservices or unrelated files to discover them.\n- Read-only inspection and validator preflight are allowed before confirmation.\n  The proposed preflight may write only the generated, non-secret `learning:`\n  patch under the repository-local `.scratch\u002F`; it is not a settings write and\n  must be removed after validation on success, failure, or cancellation. Any\n  settings-file creation or modification requires explicit confirmation.\n- Inspect an established relevant target with the **Read tool**, never `cat`. Read\n  the complete file.\n  A missing file is fine. Do not print, log, or rewrite unrelated values,\n  comments, credentials, or secret-shaped content.\n- Treat malformed or unvalidated YAML as a stop condition. Do not edit until a\n  minimal repair is understood and explicitly approved.\n- Modify only the top-level `learning:` mapping. Preserve every other byte when\n  targeted replacement permits it. Never rewrite the whole file merely to make\n  YAML easier to generate.\n- Before any write, show the complete proposed `learning:` block and exact\n  replacement diff, then ask for explicit confirmation. Silence is not consent.\n- If the file is missing, offer manual content or creation of the resolved,\n  owner-intended path. Create it only after explicit confirmation. Never add\n  credentials.\n- The validator is read-only, uses no credentials or network, and must never\n  print file contents.\n\n## Workflow\n\n### 1. Establish deployment, startup configuration, and effective owner\n\nDo this before choosing or reading a settings path. First classify the deployment\nas local standalone, local embedded mecatui, remote client\u002Fserver, or engine\nembedder. Establish the actual startup source already supplied by the operator or\ntask: service command, container args\u002Fconfig, systemd unit\u002Foverrides, or embedder\ncode. From that source record, in order:\n\n1. every repeatable `--permission-config PATH` explicit operator file;\n2. whether `--permissions-conventional` is enabled; and\n3. the server process's `XDG_CONFIG_HOME`, or its `HOME` fallback when relevant.\n\nDo not infer these from defaults when the effective invocation may override them.\nIf the startup command\u002Funit\u002Fcontainer args are not observable from authorized,\nrelevant context, ask the operator for them. Never scan running processes,\nservices, containers, or unrelated files.\n\nResolve the owning operator learning block using mecatl's actual precedence:\n\n- inspect explicit files in CLI order, stopping at the first readable, valid file\n  with a non-null top-level `learning:` block; unreadable or invalid explicit files\n  are skipped by mecatl and cannot own the effective block;\n- only when conventional discovery is enabled and no explicit file captured a\n  learning block, consider the user file at\n  `$XDG_CONFIG_HOME\u002Fmecatl\u002Fsettings.yaml`, falling back to\n  `$HOME\u002F.config\u002Fmecatl\u002Fsettings.yaml` when `XDG_CONFIG_HOME` is unset or empty;\n  its readable, valid `learning:` block is then the owner; and\n- files after the owner, and a conventional user file shadowed by an explicit\n  owner, have no effect on operator learning. Never edit one of them.\n\nApply that resolution by deployment:\n\n- **Local standalone or embedded mecatui:** resolve against the local server's\n  actual invocation and environment, not merely the interactive shell's.\n- **Remote mecatui\u002Fconnect:** resolve only from server startup args and the server\n  host's environment\u002Ffiles. Never inspect or edit local client settings. Without\n  authorized server access, provide the exact block and server-side instructions.\n- **Engine embedder:** configuration is code-owned unless the embedding\n  application explicitly constructs `permconfig` from files. Ask the owner how\n  it is supplied; do not invent a `mecated` path.\n\nIf no current block owns learning, the insertion target is the first\noperator-intended explicit file chosen by the operator, or the conventional user\nfile only when conventional discovery is enabled. If ownership or an effective\ninsertion target cannot be established, do not edit automatically: provide a\nmanual block plus instructions to install and validate it on the owning server or\nin embedder code. Record the established path privately as `\u003Cresolved-path>`.\n\n### 2. Inspect and validate the current document\n\nRead the complete target and classify it as missing, valid without `learning:`,\nvalid with `learning:`, or malformed. Retain the existing mapping, comments,\nindentation, and byte range for a later targeted edit. Report only\nlearning-related findings.\n\nOn the server host, validate the target without exposing its content. Pass the\nresolved path as one quoted argument; never concatenate it into shell syntax, a\ncommand string, or another argument:\n\n```text\nmecated config validate --file \"\u003Cresolved-path>\"\n```\n\nThe command reads at most 256 KiB and prints only `valid` or a sanitized error.\nIt requires the file to exist unless a learning patch is supplied. If `mecated`\nis unavailable on a remote host, do not auto-edit: give the exact block and\nserver-side installation guidance, including actual-file validation after the\noperator makes the change.\n\n### 3. Elicit one answer at a time\n\nAsk in this order. Wait after every question. Each prompt must show all prior\nanswers and the recommended default:\n\n```text\n✓ Autonomy: \u003Canswer>\n✓ Sensitivity: \u003Canswer>\n\n→ \u003Cone current question>\n  ▸ \u003Crecommended choice> (recommended — \u003Cbrief reason>)\n    \u003Cother choices>\n```\n\n1. **Desired autonomy:** Review first \u002F Auto \u002F Off. Recommend **Review first**\n   for rollout. If the operator explicitly asks for autonomous learning,\n   recommend **Auto** instead. Off disables observation, not explicit tools.\n2. **Sensitivity:** conservative \u002F balanced \u002F eager. Recommend **balanced**.\n   This is independent of budget posture: any sensitivity may be combined with\n   any budget profile.\n3. **Skill activation assurance:** ask this when mode is Auto. Recommend\n   **validated** for usable, body-only, evidence-backed autonomy. Offer\n   **evaluated** only when a trusted fixture evaluator is configured; it\n   requires PASS. For Review\u002FOff, record future policy or accept validated as an\n   inert default without implying activation.\n4. **Budget posture:** default \u002F budget-conscious \u002F eager \u002F custom. Profiles set\n   only the six automatic budget values; they never choose sensitivity. Explain\n   that limits and cooldowns are process-local, so N replicas may reserve about\n   N times the aggregate. They are not provider quotas or cluster-global limits.\n   For custom, ask one automatic value at a time and validate it; do not ask for\n   sensitivity again.\n5. **Existing subtree:** when present, preserve or revise it. Recommend\n   **preserve and revise only selected fields**. When absent, state that a new\n   subtree will be inserted.\n\nSupport `back` to revise the preceding answer. Do not collapse these into one\nquestionnaire.\n\n### 4. Explain the selected behavior\n\nBefore proposing YAML, summarize mode × activation accurately:\n\n- **Off:** no automatic observation\u002Freflection. Explicit `\u002Freflect`, memory and\n  user-model tools, and `SkillDraft` remain available. Direct `SkillDraft`\n  creates inactive content. Dream consolidation is independently configured.\n- **Review:** admitted completions spend reflection capacity and stage durable\n  proposals; they do not promote memory or activate learned skills.\n- **Auto:** facts retain evidence, ownership, and conflict gates. With\n  `validated`, a structurally safe, body-only, evidence-backed exact candidate\n  activates on PASS or ABSTAIN\u002Fno evaluator. With `evaluated`, only trusted\n  evaluator PASS activates. FAIL and evaluator ERROR never activate.\n\nState separately that sensitivity selects automatic admission thresholds while\nbudgets bound automatic frequency and reserved tokens; users may combine any\nsensitivity with any budget posture. Explicit `\u002Freflect` bypasses automatic\nsensitivity, cooldown, and count\u002Ftoken budgets, but retains coordinator,\nprovider, timeout, ownership, and lifecycle limits.\n\n### 5. Preflight, propose, confirm, then edit\n\nGenerate a complete `learning:` block containing mode, independently selected\nsensitivity, skill activation, and all six automatic controls. Do not emit a\npartial subtree whose behavior depends on hidden defaults.\n\nBefore asking for confirmation, write only that exact non-secret `learning:`\nblock to a bounded, repo-local scratch name such as\n`.scratch\u002Flearning-preflight.yaml`, then run the read-only in-memory preflight\nwith both paths quoted:\n\n```text\nmecated config validate --file \"\u003Cresolved-path>\" \\\n  --learning-patch \".scratch\u002Flearning-preflight.yaml\"\n```\n\nThe scratch file is not the settings write: it contains only the exact block\nalready shown, never a full settings copy, credentials, or unrelated values. The\ncommand bounded-reads both files, requires the patch to be a single YAML document\nwith exactly one top-level `learning:` mapping, replaces or inserts only that node\nin memory, and validates the resulting complete document through mecatl's parser.\nIt prints `valid` or `valid (new file)` and never writes either input. Remove the\nscratch patch immediately after validation on success, failure, or cancellation.\nStop if preflight fails.\n\nIf the user requires no scratch write, provide the complete block for manual\napplication and require actual-file validation after the write instead; do not\nclaim a write-free proposed preflight was run.\n\nThen show:\n\n1. the complete `learning:` block;\n2. an exact diff replacing only the existing mapping, or exact insertion point;\n3. a concise effects\u002Fcost summary; and\n4. `Apply this exact change to \u003Cresolved-path>? (yes\u002Fno)`.\n\nOnly explicit yes authorizes an edit. Re-read the complete target immediately\nbefore editing. If it differs from the preflight input, stop, regenerate the\nlearning-only scratch patch, rerun `mecated config validate --file\n\"\u003Cresolved-path>\" --learning-patch \".scratch\u002F\u003Cbounded-name>.yaml\"` against the new\nbytes, remove the scratch patch, and ask again. Use a targeted exact replacement\npreserving unrelated YAML and comments. If safe targeting is impossible, offer\nmanual merge rather than rewriting the file.\n\nAfter writing, validate the actual file:\n\n```text\nmecated config validate --file \"\u003Cresolved-path>\"\n```\n\nA nonzero result is a failed application requiring immediate, minimal repair\nguidance; do not report success. Re-read and verify the selected learning values\nwithout showing unrelated content.\n\n### 6. Restart and observe\n\nLearning settings are resolved at build time: restart the local server, remote\nserver, or embedder instance that owns the resolved policy.\n\n- Local embedded mecatui: `\u002Flearning` cycles mode and\n  `\u002Flearning-sensitivity` cycles sensitivity; both save locally and still\n  require restart. Use the full file for budgets and activation assurance.\n- Remote mecatui\u002Fconnect: change and restart the server host, never the client.\n- Engine embedders: follow the application's configuration\u002Frebuild lifecycle.\n- After restart, inspect `\u002Freflections`, `\u002Fskills`, and `\u002Fusermodel`; use\n  `\u002Freflect` only when an intentional live run is desired. Warn that reflection\n  may make provider calls and incur token cost; verification need not force it.\n\nOffer rollback blocks from the reference: `mode: off`, `mode: review`, and the\nhigh-assurance `skills.activation: evaluated` tightening.\n\n## Error handling\n\n| Situation | Required response |\n|---|---|\n| Startup configuration not observable or ownership unresolved | Ask for the server command\u002Funit\u002Fcontainer args; otherwise supply a manual block and server\u002Fembedder validation instructions with no automatic edit. |\n| Earlier explicit file has a valid `learning:` block | It owns learning; do not edit a later explicit or conventional file. Explicit files are evaluated in CLI order. |\n| Explicit files have no captured block and conventional discovery is disabled | The XDG user file does not participate. Ask the operator to choose an explicit insertion target or provide a manual block. |\n| Missing file | Offer manual block or confirmed creation at the established `\u003Cresolved-path>`; do not guess ownership. |\n| Malformed, duplicate, unknown, or unvalidated YAML | Stop before editing; validate the complete document and obtain approval for minimal repair. |\n| Invalid value or duration | Reject using the exact schema; `window` must be `1m..24h`. |\n| Any maximum is zero | Warn that this bound disables all automatic reflection; cooldown zero only removes cooldown. |\n| Evaluated without evaluator | Warn that ABSTAIN\u002Fno-evaluator remains staged; recommend validated or separately configure a trusted evaluator. |\n| Project raises mode\u002Fsensitivity or loosens evaluated | Explain project policy is tighten-only and cannot raise operator autonomy. |\n| Project specifies automatic budgets | Explain budgets are operator-only and the project block is warning-ignored. |\n| Multiple replicas | Multiply the process-local envelope by replica count; never call it a cluster\u002Fprovider quota. |\n| Remote server inaccessible or `mecated` unavailable there | Supply the exact manual block and server-side instructions only; never inspect local client settings or auto-edit. Require `mecated config validate --file \"\u003Cresolved-path>\"` after installation when the binary becomes available. |\n| Routing, credentials, evaluator implementation, `\u002Freflect` execution, proposal\u002Fmemory review, skill drafting, or another harness | Decline that portion and route to its workflow. |\n",{"name":5,"description":7},"main","c4ff786a7467eb2315de679ae78711d118775383",".claude\u002Fskills\u002Fmecatl-learning-config",29,"https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl","https:\u002F\u002Fmecatl.dev","https:\u002F\u002Fgithub.com\u002Fstacklok","https:\u002F\u002Fstacklok.com","Open Source AI Innovation, Grounded in Trust.","2026-11-15 04:19:54",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add stacklok\u002Fmecatl --skill mecatl-learning-config",{"tool":40,"command":41},"bunx","bunx skills add stacklok\u002Fmecatl --skill mecatl-learning-config",{"tool":43,"command":44},"pnpm","pnpm dlx skills add stacklok\u002Fmecatl --skill mecatl-learning-config","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fmecatl-learning-config\u002FSKILL.md",true,[48],{"path":49,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":52,"sourceHtmlUrl":53},"references\u002Fconfig-format.md","REFERENCE","参考文档",12750,"https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fmecatl-learning-config\u002Freferences\u002Fconfig-format.md",[55,62,69,76,83,90],{"skillNo":56,"name":57,"nameZh":6,"description":58,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M754GX1DCZR950M0SJFD048A","repo-fact","Use when asked for this repository's registered fact or codeword. Loads the canonical answer format.","https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002Fe2e\u002Ffixtures\u002Fskills\u002Frepo-fact\u002FSKILL.md",[],[],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":19},"01M754GSSV3ACG56M3XWT5F2PD","greet","Use when asked to greet someone with the ceremonial mecatl greeting. Loads the exact greeting format.","https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002Fe2e\u002Ffixtures\u002Fskills\u002Fgreet\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M754GPQN80V07DGCPCB7P2XX","go-table-tests","Write idiomatic table-driven Go tests that stay offline and follow this repo's conventions.","https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002Fdocs\u002Fexamples\u002Fskills\u002Fgo-table-tests\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M754GMQM60ARRZ0251WD6JN8","commit-style","Write Conventional-Commits messages for this repo, with the required Co-Authored-By trailer.","https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002Fdocs\u002Fexamples\u002Fskills\u002Fcommit-style\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M754GGFCQ3AWR3HHHSN3C91C","tech-writer","Use when writing or substantively editing user-facing documentation: drafting a new page, rewriting or restructuring an existing one, adding a major section, or turning engineering material (PR descriptions, specs, release notes, rough notes) into docs. Writes clear, focused technical documentation following the Diataxis framework and Mecatl's canonical style guide. Use it even when the request doesn't mention writing quality; it governs how documentation gets written. Not for editorial review of finished work.","https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ftech-writer\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M754G7VD9JBS4NVFNGREHP64","perf-optimization","Profile-driven performance optimization of mecatl using the offline benchmark + scenario harness. Use when optimizing allocations or latency, reducing allocs\u002Fop or memory, profiling a Go benchmark, pinpointing a hotspot with pprof, proving a win with benchstat, adding a regression benchmark, wiring profile-guided optimization (PGO), or investigating \"why is this slow \u002F allocating\" or a suspected perf regression. Covers task bench, task perf:scenarios, memprofile -> pprof, benchstat A\u002FB, allocs-first gating, PGO setup, and the discipline (follow the profile not the hypothesis; keep pure-perf changes byte-identical; mutation-test cache guards; skip the wrong abstraction). NOT for the live perf MCP server (use perf-mcp-interpretation) or non-mecatl Go profiling.","https:\u002F\u002Fgithub.com\u002Fstacklok\u002Fmecatl\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fperf-optimization\u002FSKILL.md",[],[]]