[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77HPVJ45G5MP2M66M6SPF3A":3,"skill-same:01M77HPVJ45G5MP2M66M6SPF3A":47},{"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":6,"updatedAt":32,"install":33,"resources":46},"01M77HPVJ45G5MP2M66M6SPF3A","trajectory-control",null,"The operator on-ramp to trajectory control (`trajctl`) — fak's live, forward-progress control plane over a DECLARED objective. Teaches the one primitive that carries the family (anything you want to progress gets a named, witnessed score, and every move is either improve-the-score or improve-the-scorer), then the four operator loops: declare an objective + plan + budget, read the CURVE (never a point) for the closed signal vocabulary (HEALTHY \u002F STALL \u002F DRIFT \u002F DETOUR_OVERRUN), apply the when-to-nudge doctrine (the REGIME GATE. Use when this named workflow matches the task.","anthony-chaudhary\u002Ffak","anthony-chaudhary","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F18080164?v=4","01M77GSKCAEKV3JS07J40B94DY","Anthony Chaudhary","01M77GSKD26Z7F4FQQD4VA30VX",48,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ftrajectory-control\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: trajectory-control\ndescription: \"The operator on-ramp to trajectory control (`trajctl`) — fak's live, forward-progress control plane over a DECLARED objective. Teaches the one primitive that carries the family (anything you want to progress gets a named, witnessed score, and every move is either improve-the-score or improve-the-scorer), then the four operator loops: declare an objective + plan + budget, read the CURVE (never a point) for the closed signal vocabulary (HEALTHY \u002F STALL \u002F DRIFT \u002F DETOUR_OVERRUN), apply the when-to-nudge doctrine (the REGIME GATE. Use when this named workflow matches the task.\"\n---\n\n# trajectory-control — declare an objective, steer by curve, nudge only by regime\n\n> **What this does.** A long-horizon session drifts: the goal from turn 1 is\n> diluted by turn 200, the context fills with operational detail, and the run\n> starts optimizing for the most recent error message instead of the objective.\n> A sub-agent adopts a nearby, more tractable goal and reports success on the\n> wrong thing. An error opens an unbounded repair side-quest. This skill is the\n> operator on-ramp to `trajctl`, fak's control plane for exactly this: give the\n> thing you want to progress a **named objective with a witnessed score**, read\n> its **curve** (not a snapshot), and intervene only when the curve — and the\n> regime — say you should. The default action is nothing.\n\nThe one primitive carries the whole family: **anything you want to progress gets\na named score**, and every conversation about progress becomes one of two moves —\n*improve the score* or *improve the scoring method*. A goal without a score is a\nwish; a score without a curve is a snapshot. **Steering reads curves, never\npoints.** The doctrine, witness rungs, steering ladder, and SOTA anchors live in\n`docs\u002Fnotes\u002FTRAJECTORY-CONTROL-SCORE-ANYTHING-2026-07-03.md` (epic **#2533**);\nthis skill is the runnable operator path over the shipped spine.\n\n---\n\n## Positioning — reach for the right trajectory skill\n\nThree skills share the word \"trajectory\"; they do not overlap.\n\n- **trajectory-control (this skill)** steers a **live** objective forward. It\n  reads a progress curve against a goal you *declared* and decides whether to\n  nudge the running session. Present tense, forward-looking.\n- **trajectory-audit** is the *retrospective* token\u002Fcost sweep of past Claude\n  Code transcripts — machine-wide I:O ratio, cache reuse, heaviest sessions,\n  stuck\u002Fchurn detectors. It scores *cost of past runs*, read-only, and never\n  touches a live objective.\n- **trajectory-garden** gardens a *recorded corpus* (`fak traj`) — near-duplicate\n  queries, cost outliers, refused traces — and proposes prunes. It scores\n  *notability of past turns*, not progress of a live goal.\n\nRule of thumb: **audit** asks \"where did the tokens go?\", **garden** asks \"which\npast turns are cruft?\", **control** asks \"is this run still going where I asked?\"\n\n`trajctl` is also **not** the scorecard family (those fold a repo surface into a\n`*_debt` integer), **not** `score-signal` (the CI-regression issue feeder), and\n**not** `loopdrive`'s binary GOAL.md witness — it adds the continuous curve\n*between* 0 and done, importing GOAL.md's Objective\u002FPlan\u002FBudget as its objective.\n\n---\n\n## Shipped surface vs. not-yet-shipped (be honest about the rung)\n\n`trajctl` is an epic mid-build. **What ships today is the data-plane leaf**\n`internal\u002Ftrajctl`: the Objective + ScoreRow model, the four witness rungs, the\nclosed status\u002Fwitness vocabularies, and the append-only JSONL ledger\n(`Append` \u002F `ReadLedgerFile` \u002F `Fold`, schema `fak-trajctl\u002F1`). It is a pure\ndata-plane leaf by design — it does not score, steer, spawn, or shell out.\n\n**Not yet shipped** (later children of #2533): the `fak trajctl declare` \u002F\n`curve` verbs, the scorer registry + built-in scorers, the stop-hook that writes\na curve point per turn, and the regime-gated re-anchor nudge. Where a step below\nwould use one of those verbs, it is marked **[NOT YET SHIPPED]** and described\nconceptually. **Do not invent output for those steps** — run only the witnessed\ncommand.\n\nThe deterministic, runnable witness of the shipped spine today:\n\n```bash\ngo test .\u002Finternal\u002Ftrajctl\u002F     # model + witness rungs + JSONL ledger round-trip, all green\n```\n\nThat test appends an objective and one score row per witness rung (W3\u002FW2\u002FW1\u002FW0),\nfolds the ledger, and proves the fold keeps the latest objective plus the full\nscore history — the ground the operator loops below stand on.\n\n---\n\n## The four operator loops\n\n### 1. Declare an objective\n\nName the thing you want to progress, its plan, and its budget. An objective is\nthe durable record `{id, parent_id, statement, plan[], scorers[], budget,\nstatus}`; the status vocabulary is closed: `active | paused | met | abandoned`.\nParent\u002Fchild ids give you the hierarchy — epic > session goal > sub-agent\nassignment > detour — so a side-quest is a *child* objective, not lost context.\n\n- **[NOT YET SHIPPED]** `fak trajctl declare` will take a GOAL.md spec (or a\n  `\u002Fgoal` condition \u002F dispatch issue) and append the objective row. Today you can\n  construct the same record through the `internal\u002Ftrajctl` API\n  (`trajctl.ObjectiveRecord` → `trajctl.Append`), which is what the green test\n  above exercises.\n- Declare the **budget** up front (turns \u002F tokens before escalation). An\n  objective with no budget cannot detect a `DETOUR_OVERRUN`.\n\n### 2. Read the curve, not the point\n\nA single score is a snapshot and lies. Steering reads the **time-ordered\nScoreRows per (objective, method)** and folds them into a closed signal\nvocabulary:\n\n| signal | what the curve shows | operator read |\n|---|---|---|\n| `HEALTHY` | rising or plateaued-at-done progress | do nothing — the default |\n| `STALL` | flat curve × high activity | the run is busy but not progressing |\n| `DRIFT` | declining alignment with the objective | it is optimizing the wrong thing |\n| `DETOUR_OVERRUN` | a detour past budget while the parent is paused | the side-quest ate the session |\n\nEvery score row carries a **witness rung**, same doctrine as `dos verify`:\n**W3** deterministic evidence (witnessed commit, green suite, benchmark) ·\n**W2** transcript\u002Fsession heuristic · **W1** judge\u002Frubric verdict · **W0**\nself-report (recorded, never gates alone). Trust the curve to the height of its\nrung — a W0 self-report is not a reason to act.\n\n- **[NOT YET SHIPPED]** `fak trajctl curve \u003Cobjective-id>` will render the fold +\n  the signal. Today the fold primitive (`trajctl.Fold` → `State.ScoresFor`) is\n  shipped and tested; the CLI rendering and the STALL\u002FDRIFT classifier are later\n  leaves.\n\n### 3. When to nudge — the regime gate\n\n**The default action is nothing.** Intervening in a high-scoring session is harm,\nnot help: mid-trajectory intervention consistently *degrades* performance in\nhigh-success regimes and helps only in low-success ones\n([arXiv:2602.03338](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.03338)). So every rung above\n\"annotate\" sits behind the **regime gate**: *recent curve healthy ⇒ do nothing.*\n\nThe steering ladder, weakest sufficient rung wins:\n\n```\nobserve → annotate (ledger only)\n        → nudge   (re-anchor: re-inject objective + curve through the session steer channel)\n        → warn    (advisory guard rung \u002F exit summary)\n        → suspend (structured refusal reason, e.g. TRAJECTORY_STALL)\n        → escalate (operator \u002F Slack)\n```\n\nRe-anchor content follows **checkpoint-and-re-read**: serialize objective + curve\n+ plan state and read it back fresh, rather than trusting context continuity\n(the most robust goal-drift mitigation in the literature —\n[2505.02709](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.02709),\n[2603.03258](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.03258)). A nudge's outcome is itself\nledgered and scored (did the curve recover after the nudge?), so the intervention\npolicy is calibrated from evidence, not vibes.\n\n- **[NOT YET SHIPPED]** the regime-gated re-anchor nudge over the `fak signal\n  steer` channel is the first steering leaf of the epic. Until it lands, apply\n  the gate **manually**: read the curve, and if it is `HEALTHY`, keep your hands\n  off the running session.\n\n### 4. Budget the detour — a side-quest is a child objective\n\nA detour is not noise to suppress — it is a **child objective with its own score\nand a budget**, and the point of trajectory control is that the detour\n*returns*. An error burst (or a correct-but-unrelated infra fix) opens a child\nobjective with a turn\u002Ftoken budget; `DETOUR_OVERRUN` fires the return-to-main\nnudge; the detour's own score records whether the repair actually landed, and the\nparent's paused time stays visible in the curve. Do not kill a recoverable\ndetour — score the recovery (the [DataPRM](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.24198)\ncorollary: scorers that punish recoverable errors prune self-correcting runs).\n\n### The meta loop — improve the score or improve the scorer\n\nScoring methods are themselves objectives. A scorer's **calibration against\nwitnessed outcomes** is scored like anything else — same shape as the scorecard\nfamily's intent-vs-literal honesty stick, pointed at the scorers. Anti-gaming\nfences are load-bearing: every metric gets an `intent_literal_scorecard` row; any\nscore that rises with session length by construction is **banned** as a steering\ninput; every scorer declares its termination condition (an unbounded scoring loop\nis a named enemy, not an accepted risk); and scores gate **harness actions only**,\nnever a model reward loop.\n\n---\n\n## The runnable demo (honest scope)\n\nThe one deterministic, model-free, network-free witness that runs end-to-end\ntoday is the data-plane test — it proves the objective\u002Fscore\u002Fwitness-rung model\nand the JSONL ledger that every loop above stands on:\n\n```bash\ngo test .\u002Finternal\u002Ftrajctl\u002F     # → ok  github.com\u002Fanthony-chaudhary\u002Ffak\u002Finternal\u002Ftrajctl\n```\n\nThe full **declare → curve → nudge** walkthrough against a live demo session\ndepends on the `fak trajctl declare` \u002F `curve` verbs and the stop-hook scorer,\nwhich are later children of epic #2533 and **not yet shipped**. When they land,\nthis section becomes a `go run .\u002Fcmd\u002Ffak trajctl …` walkthrough with a real curve;\nuntil then, do not present those verbs as runnable. See\n`docs\u002Frun-the-demos.md` for the demo entry.\n\n---\n\n## What to report\n\nA short operator note, not a wall:\n\n- the objective (statement + budget) and whether it is `active | paused | met`,\n- the **curve signal** (`HEALTHY \u002F STALL \u002F DRIFT \u002F DETOUR_OVERRUN`) and the\n  **witness rung** of the scores behind it — never quote a bare W0 as a reason,\n- the regime-gate verdict: if `HEALTHY`, the recommendation is **do nothing**,\n- for any detour, its budget, whether it overran, and whether it returned,\n- honestly, which step ran (the `go test` witness) vs. which is **not yet\n  shipped** and was described conceptually.\n\nStop there. A healthy curve is a reason to keep your hands off, not to steer.\n","# trajectory-control — declare an objective, steer by curve, nudge only by regime\n\n> **What this does.** A long-horizon session drifts: the goal from turn 1 is\n> diluted by turn 200, the context fills with operational detail, and the run\n> starts optimizing for the most recent error message instead of the objective.\n> A sub-agent adopts a nearby, more tractable goal and reports success on the\n> wrong thing. An error opens an unbounded repair side-quest. This skill is the\n> operator on-ramp to `trajctl`, fak's control plane for exactly this: give the\n> thing you want to progress a **named objective with a witnessed score**, read\n> its **curve** (not a snapshot), and intervene only when the curve — and the\n> regime — say you should. The default action is nothing.\n\nThe one primitive carries the whole family: **anything you want to progress gets\na named score**, and every conversation about progress becomes one of two moves —\n*improve the score* or *improve the scoring method*. A goal without a score is a\nwish; a score without a curve is a snapshot. **Steering reads curves, never\npoints.** The doctrine, witness rungs, steering ladder, and SOTA anchors live in\n`docs\u002Fnotes\u002FTRAJECTORY-CONTROL-SCORE-ANYTHING-2026-07-03.md` (epic **#2533**);\nthis skill is the runnable operator path over the shipped spine.\n\n---\n\n## Positioning — reach for the right trajectory skill\n\nThree skills share the word \"trajectory\"; they do not overlap.\n\n- **trajectory-control (this skill)** steers a **live** objective forward. It\n  reads a progress curve against a goal you *declared* and decides whether to\n  nudge the running session. Present tense, forward-looking.\n- **trajectory-audit** is the *retrospective* token\u002Fcost sweep of past Claude\n  Code transcripts — machine-wide I:O ratio, cache reuse, heaviest sessions,\n  stuck\u002Fchurn detectors. It scores *cost of past runs*, read-only, and never\n  touches a live objective.\n- **trajectory-garden** gardens a *recorded corpus* (`fak traj`) — near-duplicate\n  queries, cost outliers, refused traces — and proposes prunes. It scores\n  *notability of past turns*, not progress of a live goal.\n\nRule of thumb: **audit** asks \"where did the tokens go?\", **garden** asks \"which\npast turns are cruft?\", **control** asks \"is this run still going where I asked?\"\n\n`trajctl` is also **not** the scorecard family (those fold a repo surface into a\n`*_debt` integer), **not** `score-signal` (the CI-regression issue feeder), and\n**not** `loopdrive`'s binary GOAL.md witness — it adds the continuous curve\n*between* 0 and done, importing GOAL.md's Objective\u002FPlan\u002FBudget as its objective.\n\n---\n\n## Shipped surface vs. not-yet-shipped (be honest about the rung)\n\n`trajctl` is an epic mid-build. **What ships today is the data-plane leaf**\n`internal\u002Ftrajctl`: the Objective + ScoreRow model, the four witness rungs, the\nclosed status\u002Fwitness vocabularies, and the append-only JSONL ledger\n(`Append` \u002F `ReadLedgerFile` \u002F `Fold`, schema `fak-trajctl\u002F1`). It is a pure\ndata-plane leaf by design — it does not score, steer, spawn, or shell out.\n\n**Not yet shipped** (later children of #2533): the `fak trajctl declare` \u002F\n`curve` verbs, the scorer registry + built-in scorers, the stop-hook that writes\na curve point per turn, and the regime-gated re-anchor nudge. Where a step below\nwould use one of those verbs, it is marked **[NOT YET SHIPPED]** and described\nconceptually. **Do not invent output for those steps** — run only the witnessed\ncommand.\n\nThe deterministic, runnable witness of the shipped spine today:\n\n```bash\ngo test .\u002Finternal\u002Ftrajctl\u002F     # model + witness rungs + JSONL ledger round-trip, all green\n```\n\nThat test appends an objective and one score row per witness rung (W3\u002FW2\u002FW1\u002FW0),\nfolds the ledger, and proves the fold keeps the latest objective plus the full\nscore history — the ground the operator loops below stand on.\n\n---\n\n## The four operator loops\n\n### 1. Declare an objective\n\nName the thing you want to progress, its plan, and its budget. An objective is\nthe durable record `{id, parent_id, statement, plan[], scorers[], budget,\nstatus}`; the status vocabulary is closed: `active | paused | met | abandoned`.\nParent\u002Fchild ids give you the hierarchy — epic > session goal > sub-agent\nassignment > detour — so a side-quest is a *child* objective, not lost context.\n\n- **[NOT YET SHIPPED]** `fak trajctl declare` will take a GOAL.md spec (or a\n  `\u002Fgoal` condition \u002F dispatch issue) and append the objective row. Today you can\n  construct the same record through the `internal\u002Ftrajctl` API\n  (`trajctl.ObjectiveRecord` → `trajctl.Append`), which is what the green test\n  above exercises.\n- Declare the **budget** up front (turns \u002F tokens before escalation). An\n  objective with no budget cannot detect a `DETOUR_OVERRUN`.\n\n### 2. Read the curve, not the point\n\nA single score is a snapshot and lies. Steering reads the **time-ordered\nScoreRows per (objective, method)** and folds them into a closed signal\nvocabulary:\n\n| signal | what the curve shows | operator read |\n|---|---|---|\n| `HEALTHY` | rising or plateaued-at-done progress | do nothing — the default |\n| `STALL` | flat curve × high activity | the run is busy but not progressing |\n| `DRIFT` | declining alignment with the objective | it is optimizing the wrong thing |\n| `DETOUR_OVERRUN` | a detour past budget while the parent is paused | the side-quest ate the session |\n\nEvery score row carries a **witness rung**, same doctrine as `dos verify`:\n**W3** deterministic evidence (witnessed commit, green suite, benchmark) ·\n**W2** transcript\u002Fsession heuristic · **W1** judge\u002Frubric verdict · **W0**\nself-report (recorded, never gates alone). Trust the curve to the height of its\nrung — a W0 self-report is not a reason to act.\n\n- **[NOT YET SHIPPED]** `fak trajctl curve \u003Cobjective-id>` will render the fold +\n  the signal. Today the fold primitive (`trajctl.Fold` → `State.ScoresFor`) is\n  shipped and tested; the CLI rendering and the STALL\u002FDRIFT classifier are later\n  leaves.\n\n### 3. When to nudge — the regime gate\n\n**The default action is nothing.** Intervening in a high-scoring session is harm,\nnot help: mid-trajectory intervention consistently *degrades* performance in\nhigh-success regimes and helps only in low-success ones\n([arXiv:2602.03338](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.03338)). So every rung above\n\"annotate\" sits behind the **regime gate**: *recent curve healthy ⇒ do nothing.*\n\nThe steering ladder, weakest sufficient rung wins:\n\n```\nobserve → annotate (ledger only)\n        → nudge   (re-anchor: re-inject objective + curve through the session steer channel)\n        → warn    (advisory guard rung \u002F exit summary)\n        → suspend (structured refusal reason, e.g. TRAJECTORY_STALL)\n        → escalate (operator \u002F Slack)\n```\n\nRe-anchor content follows **checkpoint-and-re-read**: serialize objective + curve\n+ plan state and read it back fresh, rather than trusting context continuity\n(the most robust goal-drift mitigation in the literature —\n[2505.02709](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.02709),\n[2603.03258](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.03258)). A nudge's outcome is itself\nledgered and scored (did the curve recover after the nudge?), so the intervention\npolicy is calibrated from evidence, not vibes.\n\n- **[NOT YET SHIPPED]** the regime-gated re-anchor nudge over the `fak signal\n  steer` channel is the first steering leaf of the epic. Until it lands, apply\n  the gate **manually**: read the curve, and if it is `HEALTHY`, keep your hands\n  off the running session.\n\n### 4. Budget the detour — a side-quest is a child objective\n\nA detour is not noise to suppress — it is a **child objective with its own score\nand a budget**, and the point of trajectory control is that the detour\n*returns*. An error burst (or a correct-but-unrelated infra fix) opens a child\nobjective with a turn\u002Ftoken budget; `DETOUR_OVERRUN` fires the return-to-main\nnudge; the detour's own score records whether the repair actually landed, and the\nparent's paused time stays visible in the curve. Do not kill a recoverable\ndetour — score the recovery (the [DataPRM](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.24198)\ncorollary: scorers that punish recoverable errors prune self-correcting runs).\n\n### The meta loop — improve the score or improve the scorer\n\nScoring methods are themselves objectives. A scorer's **calibration against\nwitnessed outcomes** is scored like anything else — same shape as the scorecard\nfamily's intent-vs-literal honesty stick, pointed at the scorers. Anti-gaming\nfences are load-bearing: every metric gets an `intent_literal_scorecard` row; any\nscore that rises with session length by construction is **banned** as a steering\ninput; every scorer declares its termination condition (an unbounded scoring loop\nis a named enemy, not an accepted risk); and scores gate **harness actions only**,\nnever a model reward loop.\n\n---\n\n## The runnable demo (honest scope)\n\nThe one deterministic, model-free, network-free witness that runs end-to-end\ntoday is the data-plane test — it proves the objective\u002Fscore\u002Fwitness-rung model\nand the JSONL ledger that every loop above stands on:\n\n```bash\ngo test .\u002Finternal\u002Ftrajctl\u002F     # → ok  github.com\u002Fanthony-chaudhary\u002Ffak\u002Finternal\u002Ftrajctl\n```\n\nThe full **declare → curve → nudge** walkthrough against a live demo session\ndepends on the `fak trajctl declare` \u002F `curve` verbs and the stop-hook scorer,\nwhich are later children of epic #2533 and **not yet shipped**. When they land,\nthis section becomes a `go run .\u002Fcmd\u002Ffak trajctl …` walkthrough with a real curve;\nuntil then, do not present those verbs as runnable. See\n`docs\u002Frun-the-demos.md` for the demo entry.\n\n---\n\n## What to report\n\nA short operator note, not a wall:\n\n- the objective (statement + budget) and whether it is `active | paused | met`,\n- the **curve signal** (`HEALTHY \u002F STALL \u002F DRIFT \u002F DETOUR_OVERRUN`) and the\n  **witness rung** of the scores behind it — never quote a bare W0 as a reason,\n- the regime-gate verdict: if `HEALTHY`, the recommendation is **do nothing**,\n- for any detour, its budget, whether it overran, and whether it returned,\n- honestly, which step ran (the `go test` witness) vs. which is **not yet\n  shipped** and was described conceptually.\n\nStop there. A healthy curve is a reason to keep your hands off, not to steer.\n",{"name":5,"description":7},"main","f565b7b7db63e79649f3a02c08108132a86c5d05",".claude\u002Fskills\u002Ftrajectory-control",16,"https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak","https:\u002F\u002Fanthony-chaudhary.github.io\u002Ffak\u002Fshowcase.html","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary","https:\u002F\u002Flinkedin.com\u002Fin\u002Fanthonychaudhary","2026-11-15 03:06:00",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add anthony-chaudhary\u002Ffak --skill trajectory-control",{"tool":39,"command":40},"bunx","bunx skills add anthony-chaudhary\u002Ffak --skill trajectory-control",{"tool":42,"command":43},"pnpm","pnpm dlx skills add anthony-chaudhary\u002Ffak --skill trajectory-control","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ftrajectory-control\u002FSKILL.md",true,[],[48,55,64,71,78,83],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":52,"categories":53,"keywords":54,"publishedAt":19},"01M78V4059KRP2T88YR6Z05MJQ","serve-receipt","On-demand research protocol + deterministic receipt for serving a Hugging Face model on a concrete GPU target — fan out research lanes (model facts\u002Ffiles, engine support matrix, serving recipes, KV\u002Fstate memory math), compute the tiered feasibility verdict with `fak serve-receipt` (never narrated math), and emit a witnessed launch receipt. Use when asked for an \"initial config\", \"serving receipt\", or \"how do I serve \u003Cmodel> on \u003CGPUs>\".","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fserve-receipt\u002FSKILL.md",[],[],{"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":59,"licenseFileUrl":60,"sourceHtmlUrl":61,"categories":62,"keywords":63,"publishedAt":19},"01M77HRGKQV2CJ2K6ZFN4YQ32K","caveman","Ultra-compressed communication mode. Cuts output tokens 65% (measured) by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says \"caveman mode\", \"talk like caveman\", \"use caveman\", \"less tokens\", \"be brief\", or invokes \u002Fcaveman. Also auto-triggers when token efficiency is requested.","MIT","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002Finternal\u002Fsyspromptmmu\u002Fthird_party\u002Fcaveman\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002Finternal\u002Fsyspromptmmu\u002Fthird_party\u002Fcaveman\u002FSKILL.md",[],[],{"skillNo":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":19},"01M77HRAY3Y3BE7APN5PVJF7NV","quality-report","Score the repo quality and emit a report card.","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002Finternal\u002Fctxplans\u002Ftestdata\u002Frepo\u002F.claude\u002Fskills\u002Fquality-report\u002FSKILL.md",[],[],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":19},"01M77HR9807KHAX13DG9PW2K2Q","ctx-overlay","Compact the context window and prune the memory store when a session grows too long.","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002Finternal\u002Fctxplans\u002Ftestdata\u002Frepo\u002F.claude\u002Fskills\u002Fctx-overlay\u002FSKILL.md",[],[],{"skillNo":79,"name":66,"nameZh":6,"description":67,"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},"01M77HR7XQ9X8F2CY9Z4F2ZEZ3","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002Finternal\u002Fctxplanlint\u002Ftestdata\u002Frepo\u002F.claude\u002Fskills\u002Fquality-report\u002FSKILL.md",[],[],{"skillNo":84,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M77HR5BVKCTR1WNJGRBSDNX7","https:\u002F\u002Fgithub.com\u002Fanthony-chaudhary\u002Ffak\u002Fblob\u002Fmain\u002Finternal\u002Fctxplanlint\u002Ftestdata\u002Frepo\u002F.claude\u002Fskills\u002Fctx-overlay\u002FSKILL.md",[],[]]