[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77JD7WHKMZES4E248Z3STXP":3,"skill-same:01M77JD7WHKMZES4E248Z3STXP":49},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":13,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":28,"install":29,"resources":42},"01M77JD7WHKMZES4E248Z3STXP","doctor",null,"Diagnose Muse Code product\u002Fruntime issues from installed binary evidence. Use ONLY when the user explicitly invokes the doctor skill, asks to debug\u002Ftroubleshoot Muse Code itself, asks what happened earlier in the current Muse Code session, or explicitly selects an earlier Muse Code session. Do NOT use for ordinary repository code failures or history, benchmark tasks, implementation debugging, build\u002Ftest hangs, or third-party project issues.","raaOS\u002Fsystem_prompts_leaks","raaOS","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F214067397?v=4","01M77HV6K4ZTA4BZAGX00GTW6J","01M77HV6KX55HVRY679V25JXF8",0,"CC0-1.0","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FMeta\u002Fmuse-code\u002Fskills\u002Fdoctor\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: doctor\ndescription: Diagnose Muse Code product\u002Fruntime issues from installed binary evidence. Use ONLY when the user explicitly invokes the doctor skill, asks to debug\u002Ftroubleshoot Muse Code itself, asks what happened earlier in the current Muse Code session, or explicitly selects an earlier Muse Code session. Do NOT use for ordinary repository code failures or history, benchmark tasks, implementation debugging, build\u002Ftest hangs, or third-party project issues.\n---\n\n# Diagnose\n\nDiagnose Muse Code as an installed product. Use this skill only when the user\nexplicitly invokes `doctor` or clearly asks to debug\u002Ftroubleshoot Muse Code\nitself from binary\u002Fruntime evidence. Assume the user has the binary, not the\nsource tree. Help them understand how the app works, collect the smallest safe\nevidence set, identify the likely failing layer, and give the next safe action.\n\n## Scope\n\n- Use this skill ONLY when the user explicitly invokes `doctor`, asks to use\n  the doctor skill, or clearly asks to debug\u002Ftroubleshoot broken Muse Code\n  product behavior: app, CLI, TUI, desktop, crash, provider\u002Fmodel, settings,\n  auth, trust, skills, plugins, MCP, session, resume, export, trace, approvals,\n  sandbox, update, or unexpected output.\n- Use this skill when the user asks how Muse Code itself works, where Muse Code\n  stores state, what a Muse Code log\u002Fsession\u002Ftrace means, or how to collect a\n  Muse Code support bundle.\n- Use this skill when the user asks what happened earlier in the current\n  Muse Code session or explicitly selects an earlier Muse Code session for\n  evidence. This does not include ordinary repository history or an unspecified\n  third-party agent session.\n- Do NOT use this skill for ordinary repository engineering: code\n  implementation, third-party project bugs, benchmark\u002Feval tasks, build or test\n  failures, command hangs, toolchain issues, CI failures, or local debugging\n  inside a non-Muse Code codebase. Handle those with the normal engineering\n  workflow unless the evidence points to Muse Code itself.\n- Treat the user as a product user first, not as a repository engineer.\n- For pure settings questions or explicitly requested settings edits with no\n  product failure to investigate, use the `manage-settings` skill instead;\n  Diagnose reads settings only as evidence for a failure it is investigating.\n- Do not create issues, branches, commits, PRs, install\u002Fenable\u002Fdisable skills or\n  plugins, change settings\u002Fauth\u002Ftrust, upload logs, run live-provider\u002Fnetwork\n  checks, or edit code unless the user explicitly asks.\n- Do not print secrets, raw prompts, raw model payloads, auth tokens, API keys,\n  cookies, bearer headers, or full session logs. Prefer redacted exports,\n  key\u002Fvalue presence checks, and concise summaries.\n- If a surface has no standalone app log, say so and use session logs, crash\n  reports, trace inspection, or export evidence instead of inventing a path.\n\n## Mental Model\n\nExplain the relevant product path before asking for logs:\n\n- The binary reads settings\u002Fauth\u002Ftrust from the user's config directory and\n  writes sessions, crashes, model catalog cache, and memory under the data\n  directory.\n- A Muse Code session is the main handle for resume, trace inspection, export,\n  and support. Prefer a session id or session log path over screenshots of\n  terminal output.\n- Provider\u002Fauth failures are often config, environment, model catalog, network,\n  or credential problems. Separate those before blaming the model.\n- Skills\u002Fplugins\u002FMCP are loaded product capabilities. Diagnose discovery,\n  activation, trust, validation, and runtime errors separately.\n- A trace\u002Fexport explains what the binary saw and did. It is evidence, not a\n  transcript to paste raw.\n\n## Triage Questions\n\n1. Name the failing surface and exact symptom.\n2. Record the command, cwd, session id\u002Fpath if provided, whether the user wants\n   to inspect or continue that session, approximate time, provider\u002Fmodel if\n   relevant, and whether the issue reproduces.\n3. Ask for one missing handle only when it blocks a safe local check. Prefer:\n   exact command, session id\u002Fpath, time window, and whether they can reproduce.\n4. If the user only wants an explanation, explain first and avoid running checks.\n\n## Product Evidence Map\n\nCollect the smallest read-only set that explains the issue. Adapt the map to the\nsymptom; do not run every row by default.\n\nThe config\u002Fdata roots below are Muse Code's entire local state surface. A\nconfig, log, or session file outside them is not Muse Code state —\nnever present one as the product's active configuration, logs, or session\nevidence.\nVariables such as `CODEX_HOME` matter only for explicitly requested\nimport\u002Fcompat evidence and stay attributed to the product that owns them.\n\n1. Build\u002Fprovenance: `muse --version`; also note `command -v muse` when\n   multiple copies may exist.\n2. Config\u002Fdata roots: `$XDG_CONFIG_HOME\u002Fmuse` or `$HOME\u002F.config\u002Fmuse`;\n   `$XDG_DATA_HOME\u002Fmuse` or `$HOME\u002F.local\u002Fshare\u002Fmuse`.\n3. Local files: `settings.json`, `auth.json`, and `trust.json`; read settings\n   when relevant, but report auth\u002Ftrust presence and provider names only.\n4. Data dirs: `sessions`, `memory`, `model-catalog`, and `crashes` under the\n   data dir. For crashes, summarize report metadata and file path, not session\n   content.\n5. Session evidence: prefer `muse export --redacted --out \u003Cfile>` for the\n   latest workspace session, or `muse export --session \u003Cid-or-session.jsonl>\n   --redacted --out \u003Cfile>` when the user provides a handle.\n6. Continuation handles: if the user provides a Muse Code session id and wants\n   to continue that session, point them to `muse resume \u003Csession-id>` or\n   `muse resume --last` for interactive continuation. Use\n   `muse exec --session-id \u003Csession-id> \"\u003Cfollow-up>\"` only on explicit\n   request for headless continuation. Add `--allow-workspace-switch` to the\n   `muse exec --session-id` command only after confirming the session\n   belongs to another workspace; interactive `muse resume` does not take\n   this flag. If an exit, fork, or handoff message printed\n   `muse resume \u003Csession-id>`, treat that command as the canonical handle.\n7. Trace evidence: use `muse trace inspect --session-log \u003Csession.jsonl>\n   --render-mode compact`; add `--run-id \u003Cuuid>` or `--all-runs` for multi-run\n   logs; use `--format json` only when structured analysis is needed.\n8. User support bundle: use `\u002Ffeedback` when available; otherwise prefer a\n   redacted export, trace inspection, crash metadata, and concise reproduction\n   steps.\n9. Skills\u002Fplugins\u002FMCP: use `muse skills list --enabled-only --json` and safe\n   `muse plugins ... --help` or validation commands when the symptom points\n   there.\n10. Environment: check relevant non-secret variables by presence\u002Fvalue only, such\n   as `MUSE_MODEL`, base-url variables with credentials redacted, XDG dirs,\n   `CODEX_HOME` for import\u002Fcompat issues, and telemetry variables by presence\n   only.\n\n## Use Current Session Evidence First\n\nFor a question about what happened earlier in the current Muse Code session,\nuse the sibling `scripts\u002Fsession-evidence.py` helper before export or full trace\ninspection. After `read_skill` gives the physical Diagnose package path, run:\n\n```bash\npython3 \u003Cdoctor-skill-dir>\u002Fscripts\u002Fsession-evidence.py --session-log \u003Ccurrent-session.jsonl> --workspace \"$PWD\"\n```\n\nThe runtime session-identity context already contains the exact current log\npath. Do not ask the user for a path already present there, do not guess a\nlatest session, and do not search the session store first. A host may instead\nprovide `MUSE_CURRENT_SESSION_LOG`, in which case the helper can run without a\nselector.\n\nFor an explicitly selected earlier Muse Code session, use the exact path or id:\n\n```bash\npython3 \u003Cdoctor-skill-dir>\u002Fscripts\u002Fsession-evidence.py --session-log \u003Cexplicit-session.jsonl> --workspace \"$PWD\"\npython3 \u003Cdoctor-skill-dir>\u002Fscripts\u002Fsession-evidence.py --session-id \u003Cexplicit-session-id> --workspace \"$PWD\"\n```\n\nBoth explicit earlier-session forms are current-workspace scoped and fail\nclosed on unknown workspace metadata, a mismatch, or ambiguity. Use `--kind`,\n`--path`, `--tool`, `--run-id`, or sequence bounds to narrow follow-up evidence.\nProjected events retain their source stream, and the default bound reserves\nevidence for both the main session and child sessions so a busy child cannot\nerase the parent timeline. Always compare durable actions with assistant claims,\nespecially across compaction and child activity. Use a\nredacted export or compact trace only when this bounded projection is\ninsufficient. Never paste the raw session log into model context.\n\n## Diagnose Live Session Ownership Safely\n\nUse this path when resume says a session is already open or the original\nterminal no longer accepts input:\n\n1. Select the exact session first and run `scripts\u002Fsession-evidence.py` as\n   above. Bound the output and compare its latest durable activity timestamps;\n   do not start with a store-wide process or file search.\n2. Treat `.session.lock` as an inode-backed kernel lease, not a marker file:\n   file existence is not lock ownership, and `flock` protects an open inode.\n   Unlinking a contended pathname can let another process create and lock a new\n   inode while the original writer still owns the old inode.\n3. Probe the exact lock read-only with a non-blocking exclusive `flock`. Open it\n   without truncation, report only `acquirable`, `contended`, `missing`, or the\n   read error, then close it immediately. Never resume the session as a probe.\n4. Read the lock body's PID only as a hint. A tool sandbox or PID namespace may\n   not see the host process; `ps`\u002F`kill -0` absence inside it cannot prove that\n   the host owner died. Request a host-shell check when that distinction matters.\n5. Check the bounded session evidence for prior file mutation involving\n   `.session.lock`, especially `rm`, unlink, replacement, truncation, or\n   recreation. If the path is missing while old activity advances, stop resume\n   attempts and preserve evidence.\n\nNever remove, replace, truncate, or recreate `.session.lock` as diagnosis or\nrepair. Never signal the owner from Diagnose. Do not request takeover or use an\nowner-control endpoint. Ask the user to exit the owning process normally, then\nexplicitly retry ordinary resume.\n\nClassify the result before recommending an action:\n\n| Evidence | Classification | Safe next action |\n| --- | --- | --- |\n| Lease is acquirable | No current kernel owner; a leftover pathname is harmless | Retry normal resume; do not clean the file for tidiness |\n| contended lease with advancing session activity | Live owner and active runtime; terminal attachment may be the failed layer | Preserve the owner; inspect terminal\u002FPTY evidence or exit it normally |\n| Contended lease with bounded activity idle | Live kernel owner, runtime\u002Fterminal health unknown | Ask the user to exit the owning process normally, then explicitly retry ordinary resume |\n| Legacy owner-control artifacts are present | Diagnostic leftovers from an old binary; the kernel lease remains authoritative | Treat them as diagnostic only; exit the owning process normally before ordinary resume retry |\n| Lock path was unlinked or replaced while old activity continues | Unsafe prior mutation with possible dual writers | Stop further resume attempts, preserve both inode\u002Fsession timelines, and escalate; do not recreate the lock |\n\nA contended lease proves a live kernel owner, not that its TUI, terminal,\nprovider, or runtime is healthy. Pin the failing layer from activity plus\nhost-visible evidence before proposing a product fix.\n\n## Narrowing Loop\n\nWork from evidence:\n\n1. State the most likely layer in one or two concrete sentences.\n2. Say why the next check will confirm or reject that hypothesis, then run that\n   one safe local check.\n3. Update the hypothesis from the result and continue only while the next check\n   is still relevant and safe.\n4. If a live provider, live network, destructive command, or upload is required,\n   state why and get explicit user approval first.\n5. If customization or local state is suspected, compare against an isolated\n   temporary XDG config\u002Fdata profile only after explaining that it will not read\n   or mutate the user's real settings.\n\n## Common Diagnosis Paths\n\n- Startup\u002Fauth: binary path -> version -> config load -> auth provider present\n  -> model\u002Fcatalog selection -> first network boundary.\n- Provider\u002Fmodel: selected provider\u002Fmodel -> base URL with credentials redacted\n  -> auth presence -> model catalog cache -> trace request\u002Ferror summary.\n- Session\u002Fresume: session id\u002Fpath -> workspace match -> session log exists ->\n  resume\u002Fexport\u002Ftrace command -> whether the user wants inspection or\n  continuation.\n- Skills\u002Fplugins\u002FMCP: list\u002Fdiscovery -> activation\u002Ftrust -> validation output ->\n  runtime trace or startup diagnostic.\n- Desktop\u002FTUI: packaged app\u002Fbinary version -> session id -> UI-visible symptom\n  -> crash\u002Freport metadata -> trace\u002Fexport evidence. Say when source-only tests\n  cannot prove a packaged-app issue.\n\n## Fix Boundary\n\n- Settings-only fix: propose the exact change and apply it only after explicit\n  user request; preserve unknown settings fields and verify with a read-back or\n  focused command.\n- Workspace code fix: switch to the RED-before-GREEN engineering loop only when\n  the user explicitly asks to fix code in the current repo. Reproduce first,\n  edit surgically, rerun the same check, and report incomplete if it still\n  fails.\n- Product bug report: if the evidence points to Muse Code itself and no local fix\n  is safe, give a concise support bundle with symptom, version, config\u002Fdata\n  paths, session\u002Fcrash paths, redacted export\u002Ftrace evidence, likely cause, and\n  next action.\n\n## Completion Report\n\nInclude:\n\n- Symptom and affected surface.\n- Product mental model relevant to this failure.\n- Evidence collected, paths inspected, commands and outcomes.\n- Redactions applied.\n- Most likely cause and confidence.\n- Fix made, proposed, or not made.\n- Remaining uncertainty and the next safest check.\n","# Diagnose\n\nDiagnose Muse Code as an installed product. Use this skill only when the user\nexplicitly invokes `doctor` or clearly asks to debug\u002Ftroubleshoot Muse Code\nitself from binary\u002Fruntime evidence. Assume the user has the binary, not the\nsource tree. Help them understand how the app works, collect the smallest safe\nevidence set, identify the likely failing layer, and give the next safe action.\n\n## Scope\n\n- Use this skill ONLY when the user explicitly invokes `doctor`, asks to use\n  the doctor skill, or clearly asks to debug\u002Ftroubleshoot broken Muse Code\n  product behavior: app, CLI, TUI, desktop, crash, provider\u002Fmodel, settings,\n  auth, trust, skills, plugins, MCP, session, resume, export, trace, approvals,\n  sandbox, update, or unexpected output.\n- Use this skill when the user asks how Muse Code itself works, where Muse Code\n  stores state, what a Muse Code log\u002Fsession\u002Ftrace means, or how to collect a\n  Muse Code support bundle.\n- Use this skill when the user asks what happened earlier in the current\n  Muse Code session or explicitly selects an earlier Muse Code session for\n  evidence. This does not include ordinary repository history or an unspecified\n  third-party agent session.\n- Do NOT use this skill for ordinary repository engineering: code\n  implementation, third-party project bugs, benchmark\u002Feval tasks, build or test\n  failures, command hangs, toolchain issues, CI failures, or local debugging\n  inside a non-Muse Code codebase. Handle those with the normal engineering\n  workflow unless the evidence points to Muse Code itself.\n- Treat the user as a product user first, not as a repository engineer.\n- For pure settings questions or explicitly requested settings edits with no\n  product failure to investigate, use the `manage-settings` skill instead;\n  Diagnose reads settings only as evidence for a failure it is investigating.\n- Do not create issues, branches, commits, PRs, install\u002Fenable\u002Fdisable skills or\n  plugins, change settings\u002Fauth\u002Ftrust, upload logs, run live-provider\u002Fnetwork\n  checks, or edit code unless the user explicitly asks.\n- Do not print secrets, raw prompts, raw model payloads, auth tokens, API keys,\n  cookies, bearer headers, or full session logs. Prefer redacted exports,\n  key\u002Fvalue presence checks, and concise summaries.\n- If a surface has no standalone app log, say so and use session logs, crash\n  reports, trace inspection, or export evidence instead of inventing a path.\n\n## Mental Model\n\nExplain the relevant product path before asking for logs:\n\n- The binary reads settings\u002Fauth\u002Ftrust from the user's config directory and\n  writes sessions, crashes, model catalog cache, and memory under the data\n  directory.\n- A Muse Code session is the main handle for resume, trace inspection, export,\n  and support. Prefer a session id or session log path over screenshots of\n  terminal output.\n- Provider\u002Fauth failures are often config, environment, model catalog, network,\n  or credential problems. Separate those before blaming the model.\n- Skills\u002Fplugins\u002FMCP are loaded product capabilities. Diagnose discovery,\n  activation, trust, validation, and runtime errors separately.\n- A trace\u002Fexport explains what the binary saw and did. It is evidence, not a\n  transcript to paste raw.\n\n## Triage Questions\n\n1. Name the failing surface and exact symptom.\n2. Record the command, cwd, session id\u002Fpath if provided, whether the user wants\n   to inspect or continue that session, approximate time, provider\u002Fmodel if\n   relevant, and whether the issue reproduces.\n3. Ask for one missing handle only when it blocks a safe local check. Prefer:\n   exact command, session id\u002Fpath, time window, and whether they can reproduce.\n4. If the user only wants an explanation, explain first and avoid running checks.\n\n## Product Evidence Map\n\nCollect the smallest read-only set that explains the issue. Adapt the map to the\nsymptom; do not run every row by default.\n\nThe config\u002Fdata roots below are Muse Code's entire local state surface. A\nconfig, log, or session file outside them is not Muse Code state —\nnever present one as the product's active configuration, logs, or session\nevidence.\nVariables such as `CODEX_HOME` matter only for explicitly requested\nimport\u002Fcompat evidence and stay attributed to the product that owns them.\n\n1. Build\u002Fprovenance: `muse --version`; also note `command -v muse` when\n   multiple copies may exist.\n2. Config\u002Fdata roots: `$XDG_CONFIG_HOME\u002Fmuse` or `$HOME\u002F.config\u002Fmuse`;\n   `$XDG_DATA_HOME\u002Fmuse` or `$HOME\u002F.local\u002Fshare\u002Fmuse`.\n3. Local files: `settings.json`, `auth.json`, and `trust.json`; read settings\n   when relevant, but report auth\u002Ftrust presence and provider names only.\n4. Data dirs: `sessions`, `memory`, `model-catalog`, and `crashes` under the\n   data dir. For crashes, summarize report metadata and file path, not session\n   content.\n5. Session evidence: prefer `muse export --redacted --out \u003Cfile>` for the\n   latest workspace session, or `muse export --session \u003Cid-or-session.jsonl>\n   --redacted --out \u003Cfile>` when the user provides a handle.\n6. Continuation handles: if the user provides a Muse Code session id and wants\n   to continue that session, point them to `muse resume \u003Csession-id>` or\n   `muse resume --last` for interactive continuation. Use\n   `muse exec --session-id \u003Csession-id> \"\u003Cfollow-up>\"` only on explicit\n   request for headless continuation. Add `--allow-workspace-switch` to the\n   `muse exec --session-id` command only after confirming the session\n   belongs to another workspace; interactive `muse resume` does not take\n   this flag. If an exit, fork, or handoff message printed\n   `muse resume \u003Csession-id>`, treat that command as the canonical handle.\n7. Trace evidence: use `muse trace inspect --session-log \u003Csession.jsonl>\n   --render-mode compact`; add `--run-id \u003Cuuid>` or `--all-runs` for multi-run\n   logs; use `--format json` only when structured analysis is needed.\n8. User support bundle: use `\u002Ffeedback` when available; otherwise prefer a\n   redacted export, trace inspection, crash metadata, and concise reproduction\n   steps.\n9. Skills\u002Fplugins\u002FMCP: use `muse skills list --enabled-only --json` and safe\n   `muse plugins ... --help` or validation commands when the symptom points\n   there.\n10. Environment: check relevant non-secret variables by presence\u002Fvalue only, such\n   as `MUSE_MODEL`, base-url variables with credentials redacted, XDG dirs,\n   `CODEX_HOME` for import\u002Fcompat issues, and telemetry variables by presence\n   only.\n\n## Use Current Session Evidence First\n\nFor a question about what happened earlier in the current Muse Code session,\nuse the sibling `scripts\u002Fsession-evidence.py` helper before export or full trace\ninspection. After `read_skill` gives the physical Diagnose package path, run:\n\n```bash\npython3 \u003Cdoctor-skill-dir>\u002Fscripts\u002Fsession-evidence.py --session-log \u003Ccurrent-session.jsonl> --workspace \"$PWD\"\n```\n\nThe runtime session-identity context already contains the exact current log\npath. Do not ask the user for a path already present there, do not guess a\nlatest session, and do not search the session store first. A host may instead\nprovide `MUSE_CURRENT_SESSION_LOG`, in which case the helper can run without a\nselector.\n\nFor an explicitly selected earlier Muse Code session, use the exact path or id:\n\n```bash\npython3 \u003Cdoctor-skill-dir>\u002Fscripts\u002Fsession-evidence.py --session-log \u003Cexplicit-session.jsonl> --workspace \"$PWD\"\npython3 \u003Cdoctor-skill-dir>\u002Fscripts\u002Fsession-evidence.py --session-id \u003Cexplicit-session-id> --workspace \"$PWD\"\n```\n\nBoth explicit earlier-session forms are current-workspace scoped and fail\nclosed on unknown workspace metadata, a mismatch, or ambiguity. Use `--kind`,\n`--path`, `--tool`, `--run-id`, or sequence bounds to narrow follow-up evidence.\nProjected events retain their source stream, and the default bound reserves\nevidence for both the main session and child sessions so a busy child cannot\nerase the parent timeline. Always compare durable actions with assistant claims,\nespecially across compaction and child activity. Use a\nredacted export or compact trace only when this bounded projection is\ninsufficient. Never paste the raw session log into model context.\n\n## Diagnose Live Session Ownership Safely\n\nUse this path when resume says a session is already open or the original\nterminal no longer accepts input:\n\n1. Select the exact session first and run `scripts\u002Fsession-evidence.py` as\n   above. Bound the output and compare its latest durable activity timestamps;\n   do not start with a store-wide process or file search.\n2. Treat `.session.lock` as an inode-backed kernel lease, not a marker file:\n   file existence is not lock ownership, and `flock` protects an open inode.\n   Unlinking a contended pathname can let another process create and lock a new\n   inode while the original writer still owns the old inode.\n3. Probe the exact lock read-only with a non-blocking exclusive `flock`. Open it\n   without truncation, report only `acquirable`, `contended`, `missing`, or the\n   read error, then close it immediately. Never resume the session as a probe.\n4. Read the lock body's PID only as a hint. A tool sandbox or PID namespace may\n   not see the host process; `ps`\u002F`kill -0` absence inside it cannot prove that\n   the host owner died. Request a host-shell check when that distinction matters.\n5. Check the bounded session evidence for prior file mutation involving\n   `.session.lock`, especially `rm`, unlink, replacement, truncation, or\n   recreation. If the path is missing while old activity advances, stop resume\n   attempts and preserve evidence.\n\nNever remove, replace, truncate, or recreate `.session.lock` as diagnosis or\nrepair. Never signal the owner from Diagnose. Do not request takeover or use an\nowner-control endpoint. Ask the user to exit the owning process normally, then\nexplicitly retry ordinary resume.\n\nClassify the result before recommending an action:\n\n| Evidence | Classification | Safe next action |\n| --- | --- | --- |\n| Lease is acquirable | No current kernel owner; a leftover pathname is harmless | Retry normal resume; do not clean the file for tidiness |\n| contended lease with advancing session activity | Live owner and active runtime; terminal attachment may be the failed layer | Preserve the owner; inspect terminal\u002FPTY evidence or exit it normally |\n| Contended lease with bounded activity idle | Live kernel owner, runtime\u002Fterminal health unknown | Ask the user to exit the owning process normally, then explicitly retry ordinary resume |\n| Legacy owner-control artifacts are present | Diagnostic leftovers from an old binary; the kernel lease remains authoritative | Treat them as diagnostic only; exit the owning process normally before ordinary resume retry |\n| Lock path was unlinked or replaced while old activity continues | Unsafe prior mutation with possible dual writers | Stop further resume attempts, preserve both inode\u002Fsession timelines, and escalate; do not recreate the lock |\n\nA contended lease proves a live kernel owner, not that its TUI, terminal,\nprovider, or runtime is healthy. Pin the failing layer from activity plus\nhost-visible evidence before proposing a product fix.\n\n## Narrowing Loop\n\nWork from evidence:\n\n1. State the most likely layer in one or two concrete sentences.\n2. Say why the next check will confirm or reject that hypothesis, then run that\n   one safe local check.\n3. Update the hypothesis from the result and continue only while the next check\n   is still relevant and safe.\n4. If a live provider, live network, destructive command, or upload is required,\n   state why and get explicit user approval first.\n5. If customization or local state is suspected, compare against an isolated\n   temporary XDG config\u002Fdata profile only after explaining that it will not read\n   or mutate the user's real settings.\n\n## Common Diagnosis Paths\n\n- Startup\u002Fauth: binary path -> version -> config load -> auth provider present\n  -> model\u002Fcatalog selection -> first network boundary.\n- Provider\u002Fmodel: selected provider\u002Fmodel -> base URL with credentials redacted\n  -> auth presence -> model catalog cache -> trace request\u002Ferror summary.\n- Session\u002Fresume: session id\u002Fpath -> workspace match -> session log exists ->\n  resume\u002Fexport\u002Ftrace command -> whether the user wants inspection or\n  continuation.\n- Skills\u002Fplugins\u002FMCP: list\u002Fdiscovery -> activation\u002Ftrust -> validation output ->\n  runtime trace or startup diagnostic.\n- Desktop\u002FTUI: packaged app\u002Fbinary version -> session id -> UI-visible symptom\n  -> crash\u002Freport metadata -> trace\u002Fexport evidence. Say when source-only tests\n  cannot prove a packaged-app issue.\n\n## Fix Boundary\n\n- Settings-only fix: propose the exact change and apply it only after explicit\n  user request; preserve unknown settings fields and verify with a read-back or\n  focused command.\n- Workspace code fix: switch to the RED-before-GREEN engineering loop only when\n  the user explicitly asks to fix code in the current repo. Reproduce first,\n  edit surgically, rerun the same check, and report incomplete if it still\n  fails.\n- Product bug report: if the evidence points to Muse Code itself and no local fix\n  is safe, give a concise support bundle with symptom, version, config\u002Fdata\n  paths, session\u002Fcrash paths, redacted export\u002Ftrace evidence, likely cause, and\n  next action.\n\n## Completion Report\n\nInclude:\n\n- Symptom and affected surface.\n- Product mental model relevant to this failure.\n- Evidence collected, paths inspected, commands and outcomes.\n- Redactions applied.\n- Most likely cause and confidence.\n- Fix made, proposed, or not made.\n- Remaining uncertainty and the next safest check.\n",{"name":5,"description":7},"main","2975a5ac469614541ae22929d43b6da2635e1ee9","Meta\u002Fmuse-code\u002Fskills\u002Fdoctor","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks","https:\u002F\u002Fgithub.com\u002FraaOS","2026-11-14 21:19:17",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add raaOS\u002Fsystem_prompts_leaks --skill doctor",{"tool":35,"command":36},"bunx","bunx skills add raaOS\u002Fsystem_prompts_leaks --skill doctor",{"tool":38,"command":39},"pnpm","pnpm dlx skills add raaOS\u002Fsystem_prompts_leaks --skill doctor","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FMeta\u002Fmuse-code\u002Fskills\u002Fdoctor\u002FSKILL.md",true,[43],{"path":44,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":47,"sourceHtmlUrl":48},"scripts\u002Fsession-evidence.py","SCRIPT","脚本",31029,"https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FMeta\u002Fmuse-code\u002Fskills\u002Fdoctor\u002Fscripts\u002Fsession-evidence.py",[50,56,63,70,77,84],{"skillNo":51,"name":52,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":18},"01M77JV1M0ZZTZZREM9Q96TKZZ","model-council","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FPerplexity\u002Fskills\u002Fmodel-council\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":18},"01M77JTYND3ETS9EGCX3B9CW6E","writing-style","Draft or edit writing in the user's voice for the person who will read it and where it will appear. Use for messages, emails, documents or slides written on their behalf, not for your own replies.","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FOpenAI\u002Fdots\u002Fskills\u002Fwriting-style\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":18},"01M77JTVWEHY80JC2H1JG1JMQB","write-page","Create or edit requested Page\u002FSpace content, or use for prose you have already decided to save as a standalone Markdown file, only if the user did not explicitly request a Markdown file. Honor established formats, destinations, and repository documentation. A selected Page alone does not authorize a write.","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FOpenAI\u002Fdots\u002Fskills\u002Fwrite-page\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":18},"01M77JTQ5ND23QPF6RJYJPXHSZ","software-engineering","For use only in threads for this proactive personal assistant. Investigate software issues; inspect, write, review, or test code; work on a repository or local engineering file; or create, fix, or monitor a pull request or CI run. Use for actual engineering or product QA work, not a general programming explanation.","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FOpenAI\u002Fdots\u002Fskills\u002Fsoftware-engineering\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":18},"01M77JTM3ZDPCYE8C1FW7G7S2B","slides-artifact","Use when the user asks to create, edit, or review slides or a presentation (PowerPoint or Google Slides), including when the task clearly calls for an audience-ready slide deck. Also use to answer questions that require inspecting an existing deck; not for general presentation advice or a chat-only outline.","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FOpenAI\u002Fdots\u002Fskills\u002Fslides-artifact\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":18},"01M77JTHTWZN2J27KD2H66XDBZ","slack","Use when reading Slack conversations, deciding when to respond, or sending messages, files, and reactions in Slack.","https:\u002F\u002Fgithub.com\u002FraaOS\u002Fsystem_prompts_leaks\u002Fblob\u002Fmain\u002FOpenAI\u002Fdots\u002Fskills\u002Fslack\u002FSKILL.md",[],[]]