Explore skills
Showing 50761–50790 of 77104 skills
Use when Hugging Face model memory estimation may read, write, transmit, publish, spend, or administer data or resources. Produce a permission-boundary map for Hugging Face model memory estimation showing identity, requested capability, data scope, and approval gate. Success means every sensitive capability has a named authorization boundary, unnecessary data and privileges are called out, and no action is treated as approved by implication. The review is bounded to model weight identity, precision, sequence and batch settings, accelerator memory, and runtime overhead. Use only authorized local evidence, version-matched authoritative references, and synthetic or approved fixtures. Do not install tools, expose credentials, change live systems, publish, contact people, or make irreversible decisions without explicit approval; stop when authority or recovery is unclear.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a Hugging Face model memory estimation review is ready to be handed to a maintainer or accountable operator. Produce a release-handoff record for Hugging Face model memory estimation with version, evidence, open issues, approval state, and safe next step. Success means the receiver can distinguish verified facts, proposals, approvals, and unknowns, and no publication, deployment, write, or contact is claimed unless independently confirmed. The review is bounded to model weight identity, precision, sequence and batch settings, accelerator memory, and runtime overhead. Use only authorized local evidence, version-matched authoritative references, and synthetic or approved fixtures. Do not install tools, expose credentials, change live systems, publish, contact people, or make irreversible decisions without explicit approval; stop when authority or recovery is unclear.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a Hugging Face model memory estimation result must be independently repeated or compared without relying on an uncontrolled live action. Produce a reproducibility plan for Hugging Face model memory estimation defining fixture identity, environment, expected observations, and cleanup boundary. Success means another reviewer can identify the fixture and environment, distinguish expected from observed behavior, and repeat the check without an unapproved external effect. The review is bounded to model weight identity, precision, sequence and batch settings, accelerator memory, and runtime overhead. Use only authorized local evidence, version-matched authoritative references, and synthetic or approved fixtures. Do not install tools, expose credentials, change live systems, publish, contact people, or make irreversible decisions without explicit approval; stop when authority or recovery is unclear.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a Hugging Face model memory estimation integration or upgrade depends on compatible runtime, client, service, or artifact versions. Produce a version-compatibility record for Hugging Face model memory estimation with exact versions, supported combinations, and open questions. Success means the tested version tuple is explicit, each compatibility claim has a dated source or local result, and unsupported combinations are not presented as working. The review is bounded to model weight identity, precision, sequence and batch settings, accelerator memory, and runtime overhead. Use only authorized local evidence, version-matched authoritative references, and synthetic or approved fixtures. Do not install tools, expose credentials, change live systems, publish, contact people, or make irreversible decisions without explicit approval; stop when authority or recovery is unclear.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when operating a cloud-hosted, scheduled, or continuously running agent service beyond a single interactive session to define lifecycle controls, least-privilege scopes, budgets, observability, incident actions, and rollout gates before enabling persistent autonomy. Trigger for always-on agents, background workers, or production agent fleets.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a multi-tool workflow is nearing completion to create or update a final side-effect inventory with target, status, and verification result. Use the configured search, fetch, read, browser, and test capabilities only when available and authorized. Success criterion: each effect is either verified, explicitly pending approval, or reported as not performed. Preserve evidence and a run record, limit rework to three meaningful passes, and obtain approval before externally visible or irreversible writes.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when two or more agents, subagents, or workers must contribute to a shared objective to define bounded roles, artifact ownership, interfaces, isolated work areas, message cadence, integration checks, and a responsible coordinator. Trigger for parallel research or implementation; do not parallelize tightly coupled edits without explicit ownership and merge rules.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a project is too large for one focused implementation pass and must span multiple pull requests, sessions, or collaborating agents to create a dependency-aware plan with self-contained work packets, acceptance evidence, ownership boundaries, and change-control rules. Trigger for migrations, broad refactors, multi-stage features, or context-loss-sensitive work; skip for small tasks.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when implementing a neural-network model, training loop, inference component, or learning algorithm to define tensor shapes, data provenance, objective, and evaluation before training; verify numerical behavior on small reproducible cases before using larger or sensitive datasets. Trigger for architecture, gradient, optimizer, or model-serving work.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when multiple bounded workers produce overlapping files or findings to create or update a merged manifest of artifact, owner, source, changed paths, and checks. Use the configured search, fetch, read, browser, and test capabilities only when available and authorized. Success criterion: every accepted artifact has one accountable owner and a verified merge result. Preserve evidence and a run record, limit rework to three meaningful passes, and obtain approval before externally visible or irreversible writes.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a multi-part task cannot finish every component in one run to create or update a status matrix with dependencies and evidence per subgoal. Use the configured search, fetch, read, browser, and test capabilities only when available and authorized. Success criterion: the final status names completed, blocked, and unverified parts without contradiction. Preserve evidence and a run record, limit rework to three meaningful passes, and obtain approval before externally visible or irreversible writes.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when creating or revising a reusable agent skill for a recurring workflow to define a narrow trigger, expected inputs, ordered procedure, safety constraints, outputs, and acceptance checks; keep prose concise, separate supporting references, validate metadata and links, and check source licenses before reusing content. Trigger when instructions are repeatedly re-explained or a workflow needs consistent execution.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when work spans sessions, runs, agents, or context windows and progress must survive handoffs to record the objective, current state, decisions, evidence, unresolved risks, next action, and stopping rule in a durable artifact; verify freshness before resuming. Trigger for asynchronous, unattended, delegated, or long-running agent workflows.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a subagent returns a summary, recommendation, or code review to create or update a claim ledger with source path, tool output, and confidence. Use the configured search, fetch, read, browser, and test capabilities only when available and authorized. Success criterion: material claims can be reproduced from cited evidence or are downgraded. Preserve evidence and a run record, limit rework to three meaningful passes, and obtain approval before externally visible or irreversible writes.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a goal is too large or ambiguous to complete safely in one pass and must be broken into a sequence of verifiable tasks to identify dependencies, risks, owners, and acceptance checks; prioritize a thin path to evidence and keep scope controlled. Trigger at project starts, complex changes, or when parallel work needs a clear plan.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a agent evaluation set separation result is ready for a human owner to approve, reject, or redirect to produce a decision packet for the evaluation split-provenance record containing options, evidence, risks, and open questions. Success means the decision owner, requested decision, source evidence, alternatives, uncertainty, and consequence of no action are all visible; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a version, rule, source, or stakeholder change may affect agent evaluation set separation to produce a before/after change record and impact map for the evaluation split-provenance record. Success means each material difference is tied to affected dependencies, consumers, owners, and a test or explicitly unknown impact; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a agent evaluation set separation work item is nearing completion, pause, or transfer to another owner to produce a closeout ledger for the evaluation split-provenance record with status, evidence, owner, and retention state. Success means all deliverables, unresolved items, approvals, and next owners are recorded, and the stop reason is explicit; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when the agent evaluation set separation workflow receives records that must agree across sources, periods, or statuses to produce a reconciliation worksheet for the evaluation split-provenance record with matched, unmatched, and unresolved rows. Success means counts and key fields reconcile or every variance is quantified, sourced, and left unresolved for an owner; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when the agent evaluation set separation workflow has exceptions, missing evidence, or competing priorities to produce a prioritized exception queue for the evaluation split-provenance record with evidence, owner, and next action. Success means every exception has a severity rationale, source, owner or explicit unassigned state, and a bounded next step; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when the agent evaluation set separation workflow needs a safe recovery, rollback, replay, or continuity check to produce a recovery-drill record for the evaluation split-provenance record with starting state, test, and observed outcome. Success means the dry-run or approved non-production drill meets the predeclared recovery target and leaves the baseline intact; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when the agent evaluation set separation team needs to compare options under changing assumptions to produce a baseline-plus-scenarios matrix for the evaluation split-provenance record with assumptions and ranges. Success means the baseline is reproducible, scenario inputs are explicit, comparable units are used, and conclusions state uncertainty; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a new agent evaluation set separation request needs a bounded work scope to produce a scoped intake card for the evaluation split-provenance record. Success means owner, objective, permitted sources, acceptance condition, exclusions, and deadline are explicit; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a decision about agent evaluation set separation depends on facts from several records or public sources to produce a source-to-claim provenance ledger for the evaluation split-provenance record. Success means each material claim has a source, version/date, location, and confidence note; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a agent evaluation set separation decision depends on a limit, eligibility rule, policy, or target to produce a rule-check record for the evaluation split-provenance record showing source, units, boundary case, and outcome. Success means the rule source and effective date are verified, units and scope match, and borderline cases are flagged rather than auto-approved; held-out cases are not used to tune the tested prompt or silently relabeled. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a agent output contract validation result is ready for a human owner to approve, reject, or redirect to produce a decision packet for the output-contract fixture set containing options, evidence, risks, and open questions. Success means the decision owner, requested decision, source evidence, alternatives, uncertainty, and consequence of no action are all visible; valid and invalid cases are distinguished without inferring success from fluent prose. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a version, rule, source, or stakeholder change may affect agent output contract validation to produce a before/after change record and impact map for the output-contract fixture set. Success means each material difference is tied to affected dependencies, consumers, owners, and a test or explicitly unknown impact; valid and invalid cases are distinguished without inferring success from fluent prose. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when a agent output contract validation work item is nearing completion, pause, or transfer to another owner to produce a closeout ledger for the output-contract fixture set with status, evidence, owner, and retention state. Success means all deliverables, unresolved items, approvals, and next owners are recorded, and the stop reason is explicit; valid and invalid cases are distinguished without inferring success from fluent prose. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when the agent output contract validation workflow receives records that must agree across sources, periods, or statuses to produce a reconciliation worksheet for the output-contract fixture set with matched, unmatched, and unresolved rows. Success means counts and key fields reconcile or every variance is quantified, sourced, and left unresolved for an owner; valid and invalid cases are distinguished without inferring success from fluent prose. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodexUse when the agent output contract validation workflow has exceptions, missing evidence, or competing priorities to produce a prioritized exception queue for the output-contract fixture set with evidence, owner, and next action. Success means every exception has a severity rationale, source, owner or explicit unassigned state, and a bounded next step; valid and invalid cases are distinguished without inferring success from fluent prose. Use configured search, fetch, read, browser, test, and write capabilities only when relevant and authorized. Record evidence, limit refinement to three focused passes, and seek approval before external or irreversible actions.
Manoj-11-Dahal/try-Skills
Stars: 3Claude CodeCodex