[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CFNHMRNM3W8FY40XPD4APK":3,"skill-same:01M7CFNHMRNM3W8FY40XPD4APK":97},{"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":15,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":5,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":28,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M7CFNHMRNM3W8FY40XPD4APK","skill-miner",null,"Mine agent session logs (Claude Code, Codex, OpenCode) for user corrections, recurring failures, user frustration, token-heavy sessions, and repeated manual workflows, then find and score skills that could fix them. Use when asked \"what skills am I missing\", \"mine my session logs\", \"where do my agents keep failing\", \"find a skill for X\", \"which sessions burned the most tokens\", or to discover existing skills from skills.sh \u002F Floom \u002F local pools before writing anything new. Pairs with skill-eval-loop, which proves candidates before adoption.","getedgehq\u002Fskills","getedgehq","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F327350993?v=4","01M7CEH5YJQYF1PB3MDG9PA5HX","01M7CEH5ZAPY8TX4H3E04KM875",5,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: skill-miner\ndescription: Mine agent session logs (Claude Code, Codex, OpenCode) for user corrections, recurring failures, user frustration, token-heavy sessions, and repeated manual workflows, then find and score skills that could fix them. Use when asked \"what skills am I missing\", \"mine my session logs\", \"where do my agents keep failing\", \"find a skill for X\", \"which sessions burned the most tokens\", or to discover existing skills from skills.sh \u002F Floom \u002F local pools before writing anything new. Pairs with skill-eval-loop, which proves candidates before adoption.\n---\n\n# skill-miner\n\nTwo jobs: (1) find where the agent actually suffers in real session logs,\n(2) find the best existing or drafted skill for each pain point. Never adopt\nanything here - that is skill-eval-loop's job, with proof.\n\n## Mine corrections first (where skills win)\n\n```bash\npython3 scripts\u002Fcorrections.py --memory ~\u002F.claude\u002Fprojects\u002F\u003Cproj>\u002Fmemory --out corrections.json\n```\n\nCorrection episodes are the moments the user told the agent it got something wrong:\n(agent message, user reply). An LLM groups them into themes, marks each one\n`knowledge_gap` (fixing it needs THIS user's rules, voice, formats or facts) vs generic,\nand maps it to the user's saved rule files (`--memory`). Knowledge-gap themes sort first.\nThey are what a skill can measurably fix: frontier models already avoid generic mistakes,\nso generic-failure evals tie, while a skill carrying the user's rules wins.\n\n- Sources: Claude Code, Codex and OpenCode logs (`--sources claude,opencode,codex`, default:\n  all found). `--per-session` caps marathon sessions so one session cannot fill the budget.\n- `--no-llm` extracts episodes only (free, used by the nightly timer); `--episodes FILE`\n  clusters a saved extraction later.\n- The `session_index` also records **which skills each session loaded**, collected in the\n  same pass that reads the episodes. skill-eval-loop's recheck needs it to tell a skill\n  that earned its improvement from one that was never once loaded while the rate moved\n  anyway. Tool and argument spellings differ per source and were read off real sessions,\n  not assumed: Claude Code calls the tool `Skill` with the name in `skill`, OpenCode calls\n  it `skill` with the name in `name`, and Codex records no skill call at all.\n- The `session_index` lists **every** scanned session, not just the ones with episodes,\n  and records how many turns the human typed in each. The recheck divides by that index,\n  and an episode needs a previous user turn and a previous assistant turn, so a session\n  under two typed turns belongs in neither half of the fraction: on real logs those are\n  two thirds of all sessions, they carry no episodes, and their share of a window swings\n  wildly week to week. Without the count there is no way to drop them, and the rate ends\n  up measuring the shape of the work.\n\n## Mine tool errors, anger and spend\n\n```bash\npython3 scripts\u002Fmine.py --sources claude,opencode,codex --sessions 100 --out failures.json\n# root's history on a server: sudo python3 scripts\u002Fmine.py --projects \u002Froot\u002F.claude\u002Fprojects ...\n```\n\nOutput `failures.json`:\n- **clusters** - recurring tool errors and user corrections, classified\n  `infra` (broken environment: hooks, permissions, rate limits) vs `task`\n  (agent behavior a skill could change). Only patterns seen in >=2 sessions.\n- **anger** - user messages with frustration markers (wtf, \"why is\", \"!!!\",\n  German equivalents). Highest-signal source of workflow skills.\n- **top_spend_sessions** - sessions ranked by output tokens + error counts.\n- **repeated_workflow** clusters - near-identical opening prompts across\n  sessions. A workflow done by hand N times is a skill waiting to be written.\n\nEval-harness sessions and system banners are filtered out automatically.\n\n## Find + score candidates\n\n```bash\npython3 scripts\u002Fmatch.py failures.json --index 0 --out candidates.json\npython3 scripts\u002Fscore.py candidates.json failures.json --index 0\n```\n\n- `match.py` searches local installs, the getedgehq skills repo, the Floom\n  registry, and skills.sh (`npx skills find`, install counts included).\n- `score.py` enriches each candidate with its actual SKILL.md (fetched for\n  registry hits), the failure evidence, and ledger history, then LLM-scores\n  P(measurable improvement) 0-1. Only candidates >=0.5 deserve eval spend.\n- A candidate printed with `?` instead of a number scored nothing: the reply could\n  not be read as JSON. The reply itself is kept at\n  `$FORGE_ROOT\u002Fmined\u002Fscore-failures\u002F\u003Cname>.txt` and the line says so. Read that file\n  before changing the parser. A decoder's complaint names a column in a reply nobody\n  kept, and a parser fixed from the complaint alone is fixed for a guess.\n- Nothing matches? `draft.py \u003Cclusters.json> --index 0` drafts a minimal candidate\n  skill into `$FORGE_ROOT\u002Fdrafts\u002F`. For correction themes it automatically feeds the\n  mapped memory files and the user's verbatim rules as enriched context; add more with\n  `--context FILE` (accepted examples, past eval losses). Generic best practice never\n  beats a frontier model, so a draft that quotes no user rule is a weak candidate.\n\n## Check the scorer against what the evals decided\n\n```bash\npython3 scripts\u002Fcalibrate.py            # adoption rates per bucket, or \"too few to call\"\npython3 scripts\u002Fcalibrate.py --check    # did high scores actually predict adoption?\n```\n\n`calibrate.py` reads the ledger and reports the adoption rate per bucket - failure\nkind, skill provenance, retries of a skill that already lost - and any bucket under\nfour decided evals comes out as \"too few to call\" rather than as a rate. A 1-of-1\nbucket is noise, and a prior stated with false confidence is worse than no prior.\n\nBoth ends of that are now wired up, so the prediction is checkable instead of\ndecorative:\n\n- **The priors reach the prompt.** `score.py` pastes `calibrate.py`'s measured block\n  into its own scoring prompt, so the scorer reasons from this user's base rates\n  instead of a general impression of what a good skill looks like. Under eight decided\n  evals in total it states that there are too few and scores from the evidence alone.\n- **Every prediction is written down.** Each scored candidate appends a row to\n  `predictions.jsonl`, which is what `--check` joins to the decisions. A failed model\n  call records nothing: a failure is not a prediction of 0, it is no prediction.\n- **And an unreadable reply is a failed call, not a zero.** Reading the verdict by\n  slicing from the first `{` to the last `}` breaks on a model that answers with the\n  object and then keeps going: the slice is valid JSON followed by more, `json.loads`\n  rejects all of it, and two real candidates scored 0.00 in a live pass for a judgement\n  the model never made. The reply is now parsed to the end of the first complete object,\n  and a candidate that still cannot be read is carried as unscored and sorts last,\n  rather than being quietly retired at the 0.5 threshold.\n- **`--check` reports coverage, not just the gap.** A report that pairs one prediction\n  reads the same whether the ledger is young or the scorer is broken, and on this one it\n  read that way for weeks: 1 of 19 decided evals had a prediction, and the missing 18 were\n  a crash, not a wait. `--check` now prints the fraction and names the decisions with no\n  prediction, so the next systematic gap shows up as a number in the same pass that reads\n  the gap rather than six weeks later.\n- **A brief nobody mined still gets a prediction.** `--skill-dir` reads the cluster out of\n  the brief's `source_failure`, which only `draft.py` writes. A brief written by hand has\n  no such key, the fallback handed the brief itself back as a cluster, and the next line\n  raised `KeyError: 'signature'`. `forge.sh` prints that as \"scoring failed, evaluating\n  anyway\", which reads like one unlucky model call - but 22 of the 37 briefs here have no\n  `source_failure`, so the prediction log added to close the prediction-to-decision gap was\n  never going to see the majority of decisions, the first run against a real incumbent\n  among them. Such a brief now scores against its own task, and the prompt says the missing\n  cluster is no evidence either way rather than letting rubric item 3 read an absent\n  history as a weak one.\n- **Provenance is a class, not a path.** The ledger stores `skill_src` as a filesystem\n  path, so bucketing it raw produced one n=1 bucket per draft - a dimension that could\n  never say anything. Paths now class into drafted by the loop \u002F already installed\n  locally \u002F found in a registry.\n- **The prediction is about the skill that gets evaluated.** Writing predictions down\n  is worth nothing if they are about other skills. `score.py` only ever saw the\n  candidates `match.py` found, and a matched candidate is used only when it scores at\n  least 0.5, so the skill that reached the gate was nearly always a draft nobody had\n  scored: 8 predictions, 21 decisions, and no pair between them however many evals ran.\n  That is not a wiring bug to wait out, it is a design that could not close. `score.py\n  \u003Cbrief> --skill-dir \u003Cdir>` scores exactly the skill about to be run, under the\n  basename the ledger records it by, from the cluster the brief already carries in\n  `source_failure`. `forge.sh` calls it once per eval, so every decision from here on\n  has one prediction to pair with. The prediction is recorded, not acted on: skipping\n  an eval on a low score needs `--check` to first say these numbers separate winners\n  from losers, which is the question this makes askable.\n\nOn the live ledger that turns sixteen unusable buckets into one real number: skills\nthe loop drafted itself pass the gate 9 of 16 (56%), corrections 7 of 10 (70%),\nunlabelled failures 3 of 5 (60%); retries, tool errors and registry finds all still\nreport as too few to call. `tests\u002Ftest_calibrate.py` covers both halves on fixtures -\nstdlib only, no model calls - including that a thin ledger states no priors at all and\nthat an unwritable prediction log never stops a scoring pass.\n\n## Rules\n\n- Infra-class clusters go to **agent-infra-fixer** (fix the hook\u002Fpermission\u002Fwrapper,\n  verify, record in ledger). No skill fixes a broken environment.\n- Drafted skills must name task-context triggers in their description -\n  \"prevents X error\" never gets loaded, the agent can't know it will err.\n- A candidate is a hypothesis, not an adoption. Hand off to skill-eval-loop.\n\nState lives in `$FORGE_ROOT` (default `~\u002Fskill-forge`): mined\u002F, drafts\u002F, ledger.jsonl.\n","# skill-miner\n\nTwo jobs: (1) find where the agent actually suffers in real session logs,\n(2) find the best existing or drafted skill for each pain point. Never adopt\nanything here - that is skill-eval-loop's job, with proof.\n\n## Mine corrections first (where skills win)\n\n```bash\npython3 scripts\u002Fcorrections.py --memory ~\u002F.claude\u002Fprojects\u002F\u003Cproj>\u002Fmemory --out corrections.json\n```\n\nCorrection episodes are the moments the user told the agent it got something wrong:\n(agent message, user reply). An LLM groups them into themes, marks each one\n`knowledge_gap` (fixing it needs THIS user's rules, voice, formats or facts) vs generic,\nand maps it to the user's saved rule files (`--memory`). Knowledge-gap themes sort first.\nThey are what a skill can measurably fix: frontier models already avoid generic mistakes,\nso generic-failure evals tie, while a skill carrying the user's rules wins.\n\n- Sources: Claude Code, Codex and OpenCode logs (`--sources claude,opencode,codex`, default:\n  all found). `--per-session` caps marathon sessions so one session cannot fill the budget.\n- `--no-llm` extracts episodes only (free, used by the nightly timer); `--episodes FILE`\n  clusters a saved extraction later.\n- The `session_index` also records **which skills each session loaded**, collected in the\n  same pass that reads the episodes. skill-eval-loop's recheck needs it to tell a skill\n  that earned its improvement from one that was never once loaded while the rate moved\n  anyway. Tool and argument spellings differ per source and were read off real sessions,\n  not assumed: Claude Code calls the tool `Skill` with the name in `skill`, OpenCode calls\n  it `skill` with the name in `name`, and Codex records no skill call at all.\n- The `session_index` lists **every** scanned session, not just the ones with episodes,\n  and records how many turns the human typed in each. The recheck divides by that index,\n  and an episode needs a previous user turn and a previous assistant turn, so a session\n  under two typed turns belongs in neither half of the fraction: on real logs those are\n  two thirds of all sessions, they carry no episodes, and their share of a window swings\n  wildly week to week. Without the count there is no way to drop them, and the rate ends\n  up measuring the shape of the work.\n\n## Mine tool errors, anger and spend\n\n```bash\npython3 scripts\u002Fmine.py --sources claude,opencode,codex --sessions 100 --out failures.json\n# root's history on a server: sudo python3 scripts\u002Fmine.py --projects \u002Froot\u002F.claude\u002Fprojects ...\n```\n\nOutput `failures.json`:\n- **clusters** - recurring tool errors and user corrections, classified\n  `infra` (broken environment: hooks, permissions, rate limits) vs `task`\n  (agent behavior a skill could change). Only patterns seen in >=2 sessions.\n- **anger** - user messages with frustration markers (wtf, \"why is\", \"!!!\",\n  German equivalents). Highest-signal source of workflow skills.\n- **top_spend_sessions** - sessions ranked by output tokens + error counts.\n- **repeated_workflow** clusters - near-identical opening prompts across\n  sessions. A workflow done by hand N times is a skill waiting to be written.\n\nEval-harness sessions and system banners are filtered out automatically.\n\n## Find + score candidates\n\n```bash\npython3 scripts\u002Fmatch.py failures.json --index 0 --out candidates.json\npython3 scripts\u002Fscore.py candidates.json failures.json --index 0\n```\n\n- `match.py` searches local installs, the getedgehq skills repo, the Floom\n  registry, and skills.sh (`npx skills find`, install counts included).\n- `score.py` enriches each candidate with its actual SKILL.md (fetched for\n  registry hits), the failure evidence, and ledger history, then LLM-scores\n  P(measurable improvement) 0-1. Only candidates >=0.5 deserve eval spend.\n- A candidate printed with `?` instead of a number scored nothing: the reply could\n  not be read as JSON. The reply itself is kept at\n  `$FORGE_ROOT\u002Fmined\u002Fscore-failures\u002F\u003Cname>.txt` and the line says so. Read that file\n  before changing the parser. A decoder's complaint names a column in a reply nobody\n  kept, and a parser fixed from the complaint alone is fixed for a guess.\n- Nothing matches? `draft.py \u003Cclusters.json> --index 0` drafts a minimal candidate\n  skill into `$FORGE_ROOT\u002Fdrafts\u002F`. For correction themes it automatically feeds the\n  mapped memory files and the user's verbatim rules as enriched context; add more with\n  `--context FILE` (accepted examples, past eval losses). Generic best practice never\n  beats a frontier model, so a draft that quotes no user rule is a weak candidate.\n\n## Check the scorer against what the evals decided\n\n```bash\npython3 scripts\u002Fcalibrate.py            # adoption rates per bucket, or \"too few to call\"\npython3 scripts\u002Fcalibrate.py --check    # did high scores actually predict adoption?\n```\n\n`calibrate.py` reads the ledger and reports the adoption rate per bucket - failure\nkind, skill provenance, retries of a skill that already lost - and any bucket under\nfour decided evals comes out as \"too few to call\" rather than as a rate. A 1-of-1\nbucket is noise, and a prior stated with false confidence is worse than no prior.\n\nBoth ends of that are now wired up, so the prediction is checkable instead of\ndecorative:\n\n- **The priors reach the prompt.** `score.py` pastes `calibrate.py`'s measured block\n  into its own scoring prompt, so the scorer reasons from this user's base rates\n  instead of a general impression of what a good skill looks like. Under eight decided\n  evals in total it states that there are too few and scores from the evidence alone.\n- **Every prediction is written down.** Each scored candidate appends a row to\n  `predictions.jsonl`, which is what `--check` joins to the decisions. A failed model\n  call records nothing: a failure is not a prediction of 0, it is no prediction.\n- **And an unreadable reply is a failed call, not a zero.** Reading the verdict by\n  slicing from the first `{` to the last `}` breaks on a model that answers with the\n  object and then keeps going: the slice is valid JSON followed by more, `json.loads`\n  rejects all of it, and two real candidates scored 0.00 in a live pass for a judgement\n  the model never made. The reply is now parsed to the end of the first complete object,\n  and a candidate that still cannot be read is carried as unscored and sorts last,\n  rather than being quietly retired at the 0.5 threshold.\n- **`--check` reports coverage, not just the gap.** A report that pairs one prediction\n  reads the same whether the ledger is young or the scorer is broken, and on this one it\n  read that way for weeks: 1 of 19 decided evals had a prediction, and the missing 18 were\n  a crash, not a wait. `--check` now prints the fraction and names the decisions with no\n  prediction, so the next systematic gap shows up as a number in the same pass that reads\n  the gap rather than six weeks later.\n- **A brief nobody mined still gets a prediction.** `--skill-dir` reads the cluster out of\n  the brief's `source_failure`, which only `draft.py` writes. A brief written by hand has\n  no such key, the fallback handed the brief itself back as a cluster, and the next line\n  raised `KeyError: 'signature'`. `forge.sh` prints that as \"scoring failed, evaluating\n  anyway\", which reads like one unlucky model call - but 22 of the 37 briefs here have no\n  `source_failure`, so the prediction log added to close the prediction-to-decision gap was\n  never going to see the majority of decisions, the first run against a real incumbent\n  among them. Such a brief now scores against its own task, and the prompt says the missing\n  cluster is no evidence either way rather than letting rubric item 3 read an absent\n  history as a weak one.\n- **Provenance is a class, not a path.** The ledger stores `skill_src` as a filesystem\n  path, so bucketing it raw produced one n=1 bucket per draft - a dimension that could\n  never say anything. Paths now class into drafted by the loop \u002F already installed\n  locally \u002F found in a registry.\n- **The prediction is about the skill that gets evaluated.** Writing predictions down\n  is worth nothing if they are about other skills. `score.py` only ever saw the\n  candidates `match.py` found, and a matched candidate is used only when it scores at\n  least 0.5, so the skill that reached the gate was nearly always a draft nobody had\n  scored: 8 predictions, 21 decisions, and no pair between them however many evals ran.\n  That is not a wiring bug to wait out, it is a design that could not close. `score.py\n  \u003Cbrief> --skill-dir \u003Cdir>` scores exactly the skill about to be run, under the\n  basename the ledger records it by, from the cluster the brief already carries in\n  `source_failure`. `forge.sh` calls it once per eval, so every decision from here on\n  has one prediction to pair with. The prediction is recorded, not acted on: skipping\n  an eval on a low score needs `--check` to first say these numbers separate winners\n  from losers, which is the question this makes askable.\n\nOn the live ledger that turns sixteen unusable buckets into one real number: skills\nthe loop drafted itself pass the gate 9 of 16 (56%), corrections 7 of 10 (70%),\nunlabelled failures 3 of 5 (60%); retries, tool errors and registry finds all still\nreport as too few to call. `tests\u002Ftest_calibrate.py` covers both halves on fixtures -\nstdlib only, no model calls - including that a thin ledger states no priors at all and\nthat an unwritable prediction log never stops a scoring pass.\n\n## Rules\n\n- Infra-class clusters go to **agent-infra-fixer** (fix the hook\u002Fpermission\u002Fwrapper,\n  verify, record in ledger). No skill fixes a broken environment.\n- Drafted skills must name task-context triggers in their description -\n  \"prevents X error\" never gets loaded, the agent can't know it will err.\n- A candidate is a hypothesis, not an adoption. Hand off to skill-eval-loop.\n\nState lives in `$FORGE_ROOT` (default `~\u002Fskill-forge`): mined\u002F, drafts\u002F, ledger.jsonl.\n",{"name":5,"description":7},"main","01960aeb08909dbd0c1a2d4f219cdb300099b8b9",0,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills","https:\u002F\u002Fgetedge.cc","https:\u002F\u002Fgithub.com\u002Fgetedgehq","2026-11-12 10:05:02",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add getedgehq\u002Fskills --skill skill-miner",{"tool":37,"command":38},"bunx","bunx skills add getedgehq\u002Fskills --skill skill-miner",{"tool":40,"command":41},"pnpm","pnpm dlx skills add getedgehq\u002Fskills --skill skill-miner","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002FSKILL.md",true,[45,51,55,61,65,69,73,77,81,85,89,93],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"DERIVATION.json","REFERENCE","参考文档",2962,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002FDERIVATION.json",{"path":52,"resourceType":52,"resourceTypeDesc":53,"sizeBytes":54,"sourceHtmlUrl":15},"LICENSE","许可证",10756,{"path":56,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":59,"sourceHtmlUrl":60},"scripts\u002Fcalibrate.py","SCRIPT","脚本",8287,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fcalibrate.py",{"path":62,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":63,"sourceHtmlUrl":64},"scripts\u002Fcorrections.py",12601,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fcorrections.py",{"path":66,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":67,"sourceHtmlUrl":68},"scripts\u002Fdraft.py",5395,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fdraft.py",{"path":70,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":71,"sourceHtmlUrl":72},"scripts\u002Fforge_llm.py",5341,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fforge_llm.py",{"path":74,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":75,"sourceHtmlUrl":76},"scripts\u002Fmatch.py",5257,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fmatch.py",{"path":78,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":79,"sourceHtmlUrl":80},"scripts\u002Fmine.py",10217,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fmine.py",{"path":82,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":83,"sourceHtmlUrl":84},"scripts\u002Fscore.py",14564,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fscore.py",{"path":86,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":87,"sourceHtmlUrl":88},"scripts\u002Fsources.py",17188,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Fscripts\u002Fsources.py",{"path":90,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":91,"sourceHtmlUrl":92},"tests\u002Ftest_calibrate.py",18227,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Ftests\u002Ftest_calibrate.py",{"path":94,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":95,"sourceHtmlUrl":96},"tests\u002Ftest_corrections.py",2913,"https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fskill-miner\u002Ftests\u002Ftest_corrections.py",[98,106,114,121,128,136],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":102,"sourceHtmlUrl":103,"categories":104,"keywords":105,"publishedAt":19},"01M7CFVPC1P7JN60J5AN9AZZ5G","zero-downtime-cloud-move","Runbook for moving a live app from one cloud to another (Google Cloud to AWS, AWS to Google Cloud, a PaaS to your own account, or one storage provider to another) without downtime. The agent inventories what is really running, builds the new stack next to the old one, keeps a one-line rollback for every phase, proves the new stack with real workloads before any traffic moves, cuts over with a checklist, watches a fixed window after the switch, copies data with an idempotent job whose second pass must report 0 left, and deletes the old cloud last. Use when someone asks to migrate, move or replatform a production app, switch cloud providers, move a backend, database files or object storage, plan a DNS cutover, or decommission an old cloud account.","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fzero-downtime-cloud-move\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fzero-downtime-cloud-move\u002FSKILL.md",[],[],{"skillNo":107,"name":108,"nameZh":6,"description":109,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":110,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":19},"01M7CFVH8VN6G2KE8PBFD1CRMN","zayats-ceo-branding","Build an evidence-based CEO positioning paper, Personal Branding Roadmap, and editorial workflow using Marina Zayats's public method. Use for executive branding, LinkedIn ghostwriting briefs, profile strategy, and recurring CEO-approved content planning.","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fzayats-ceo-branding\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fzayats-ceo-branding\u002FSKILL.md",[],[],{"skillNo":115,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":118,"categories":119,"keywords":120,"publishedAt":19},"01M7CFV9BB7G8K8CX9RKZ3GECX","yc-app-roast","YC App Roast: roast a YC or a16z Speedrun application. Use when a founder wants their YC or Speedrun answers scored, asks \"what would a YC partner think\", wants the weak answers found or rewritten, or wants their draft checked before they submit. Scores every answer 0-10 against public advice from YC and Speedrun partners, names the top issue with the exact words that cost points, and writes a rewrite from the founder's own facts. Works for YC and a16z Speedrun. Not affiliated with Y Combinator or a16z.","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fyc-app-roast\u002FSKILL.md",[],[],{"skillNo":122,"name":123,"nameZh":6,"description":124,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":125,"categories":126,"keywords":127,"publishedAt":19},"01M7CFV11Z1HB9EEWFQ6PCJ98X","yc-app-review","YC App Review: get a YC or a16z Speedrun application reviewed by people you trust. Use when a founder wants friends, mentors or alumni to review their YC or Speedrun application, or asks to share their draft for feedback. Builds a shareable review page (one HTML file, application-portal look) where reviewers leave notes per question and press \"Send my notes\" to send them back as text, then parses the notes into proposed rewrites. Optional tracked private links per reviewer. Works for YC and a16z Speedrun. Not affiliated with Y Combinator or a16z.","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fyc-app-review\u002FSKILL.md",[],[],{"skillNo":129,"name":130,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":132,"sourceHtmlUrl":133,"categories":134,"keywords":135,"publishedAt":19},"01M7CFTSVJV7XCSW62W39MX8CZ","x-growth-playbook","X Growth Playbook. Analyse a person's own X (Twitter) posts and tell them what to post next, with the numbers behind every claim. Compares quote posts vs originals, links vs no links, short vs long, time of day in their timezone, media vs text and the topics they name, reports medians with n, then writes a 10-step plan and the next 7 posts in their voice. Use when the user says \"analyse my X posts\", \"what works on my Twitter\", \"how do I grow on X\", \"what should I post next on X\", \"audit my X account\", or shares an X data archive, X API export or a list of their posts. Runs on the user's computer.","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fx-growth-playbook\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fx-growth-playbook\u002FSKILL.md",[],[],{"skillNo":137,"name":138,"nameZh":6,"description":139,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":140,"sourceHtmlUrl":141,"categories":142,"keywords":143,"publishedAt":19},"01M7CFTK101S3P00MQ129G5YNH","workplan","Create, update, or close work plans for multi-step tasks. Use when starting refactors, bug lists, feature work, migrations, or any task with 2+ steps. Also use after auto-compaction to re-orient. Triggers: \"workplan\", \"work plan\", \"create a plan\", \"what's the plan\", \"where was I\", or when Claude detects a multi-step task that needs tracking.","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fworkplan\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fgetedgehq\u002Fskills\u002Fblob\u002Fmain\u002Fworkplan\u002FSKILL.md",[],[]]