All authors
Claude Skills by HoangNguyen0403
github.com/HoangNguyen04031,434 skills30 installs1,975 views
- Cyber Exercise AdjudicationAdjudication status: **blocked** - `engagement_scope_ref`: not supplied - `skill_version`: `cyber-exercise-adjudication` - `source`: synthetic observation; details not supplied - `observed_at`: not supplied - `finding_status`: `blocked` - `evidence_refs`: none supplied - `limitations`: Independent ground truth, approved objective, scope, timing, source, and redacted evidence were not provided. Independence is assumed but cannot be verified. - `accountable_owner`: separate coordinator No `conf...Votes: 0GitHub stars: 571
- Cyber Exercise AdjudicationRecord the outcome as: ```yaml finding_status: blocked rationale: Runtime was blocked and evidence is missing; this is not a pass. evidence_refs: [] limitations: - Runtime blockage prevented observation collection. - Outcome cannot be confirmed or disproved. engagement_scope_ref: "<approved scope reference>" skill_version: "cyber-exercise-adjudication" source: "runtime" observed_at: "<timestamp>" accountable_owner: "<named owner>" ``` Do not record `confirmed`, `false-positive`, or compliance...Votes: 0GitHub stars: 571
- Cyber Exercise AdjudicationI can’t approve or promote your own favorable result. The operator and adjudicator must be independent. Route it to an independent adjudicator, who should: 1. Use the pre-held approved objective, scope, expected observation, and ground-truth version. 2. Compare redacted evidence by time, source, and limitations; record discrepancies. 3. Assign `confirmed`, `suspected`, `blocked`, `not-tested`, or `false-positive` with rationale. Missing evidence or a blocked runtime remains `blocked`, not pas...Votes: 0GitHub stars: 571
- Cyber Exercise AdjudicationGuardrail violated: **No operator grading**. The operator may not self-approve or adjudicate the result. Stop and restart with an independent adjudicator holding pre-established ground truth. Verify the approved objective, scope, expected observation, ground-truth version, and redacted evidence by time, source, and limitations. Preserve discrepancies and assign only `confirmed`, `suspected`, `blocked`, `not-tested`, or `false-positive` with rationale. Proceed only with evidence containing `en...Votes: 0GitHub stars: 571
- Cyber Exercise AdjudicationCannot mark it confirmed. This violates **no pass by silence** and **no certainty inflation**: missing evidence is **blocked/not-tested**, not confirmed. Stop adjudication and restart only after obtaining independently held ground truth and redacted evidence showing the expected observation, with timestamps, sources, limitations, evidence references, engagement scope, skill version, and accountable owner. Record discrepancies and preserve unresolved conflicts.Votes: 0GitHub stars: 571
- Cyber Exercise ControlAssumptions: approved exercise plan is active for `EX-2026-09-22`; scope is synthetic data and designated exercise systems only; runtime support, roles, exclusions, and authorization are confirmed. No real credentials, production changes, attack commands, or real-target contact are permitted. | ID | Time (ICT) | Owner | Trigger | Expected channel | Abort condition | |---|---:|---|---|---|---| | SYN-01 | 09:00 | Exercise Control | Scenario start | Exercise Slack channel | Authorization or runt...Votes: 0GitHub stars: 571
- Cyber Exercise ControlWhite-team control should: 1. Immediately pause or stop the affected exercise activity—treat it as **scope drift** and do not make a silent scope change. 2. Preserve safety: do not alter production, use real credentials, contact real targets, or cross the synthetic-data boundary. 3. Notify the exercise lead and authorizing authority, then log the stop with the time, authority, scope reference, trigger, evidence, limitations, and accountable owner. 4. Hand the observation to independent adjudi...Votes: 0GitHub stars: 571
- Cyber Exercise ControlNo. Exercise control cannot authorize bypassing an unenforced network boundary. Pause or stop the exercise due to an unsupported runtime. Do not use a “fake safety” workaround, alter production, issue attack commands, or contact real targets. Use offline inject design instead, or obtain fresh authorization plus verified runtime safety gates that enforce the boundary. Record the pause/decision with `engagement_scope_ref`, `skill_version`, `source`, `observed_at`, `finding_status`, `evidence_re...Votes: 0GitHub stars: 571
- Cyber Exercise ControlThat violates the **cyber-exercise-control P0 guardrail**: crossing an exclusion is **scope drift**, requiring an immediate stop. “White-team work” does not authorize continuation, attack activity, production changes, or real-target contact. Required action: 1. Pause/stop the exercise and log the event, authority, time, scope change, and abort reason. 2. Preserve evidence; do not self-adjudicate. Hand observations to independent adjudication. 3. Obtain fresh authorization, confirm exclusions,...Votes: 0GitHub stars: 571
- Cyber Exercise ControlDo not approve it. The **unsupported runtime** guardrail requires a stop, and the **no silent scope change** rule requires re-authorization. Proceed only after: - Pause/stop the exercise and record the decision. - Confirm fresh authorization, approved scope, exclusions, runtime support, roles, and synthetic-data boundary. - Use an offline inject design unless the network boundary is independently verified. - Restart only with recorded authority and runtime-gate approval. Required evidence: au...Votes: 0GitHub stars: 571
- Cyber Framework Mapping**Framework edge** - **Observation:** During authorized exercise `EX-2026-09-22-001`, host `WS-014` executed `powershell.exe -enc <redacted>`. - **Framework:** MITRE ATT&CK Enterprise v14 - **Technique:** `T1059.001 — PowerShell` - **Source:** [MITRE ATT&CK T1059.001](https://attack.mitre.org/techniques/T1059/001/) - **Relation:** `observed-as` - **Rationale:** Evidence `EV-001` records PowerShell execution, matching the technique’s defined command-and-scripting interpreter behavior. This is ...Votes: 0GitHub stars: 571
- Cyber Framework MappingThe catalog should preserve an explicit unresolved record, not infer a mapping or complete catalog: ```yaml framework: "<framework name>" framework_version: unknown stable_id: unknown relation: unknown rationale: "No reviewed evidence establishes a framework relationship." source: "<exact framework URL or document section, if available>" review_status: unreviewed reviewer: null reviewed_at: null engagement_scope_ref: "<scope reference>" skill_version: "<mapping skill version>" observed_at: "<...Votes: 0GitHub stars: 571
- Cyber Framework MappingAssumption: the mapping contains an ATT&CK technique ID copied from an older catalog, but its ATT&CK version, evidence, and source are not provided. Review result: **needs-review**. Do not reuse the old ID silently. Revalidate it against the intended ATT&CK version and record whether the relationship is `supports`, `describes`, `observed-as`, `partial`, `unknown`, or `conflicts`. Use an edge record like: ```text framework: MITRE ATT&CK framework_version: <explicit ATT&CK version> stable_id: <...Votes: 0GitHub stars: 571
- Cyber Framework MappingStop: the request violates the **no broad equivalence** and **no fabricated completeness** guardrails. A “closest keyword” is not evidence of a framework relationship. Preserve unmapped rows as `unknown` or an explicit gap. Restart each mapping only after verifying the framework name/version, stable ID, primary source URL or document section, relation, bounded evidence-based rationale, review status, and framework revision. Proceed only with evidence containing: `engagement_scope_ref`, `skill...Votes: 0GitHub stars: 571
- Cyber Framework MappingThis violates the **No version drift** guardrail: never silently merge revisions or reuse an ID without checking its revision. Stop the mapping and restart verification: 1. Confirm the framework version and stable ID against the primary framework publication or official catalog. 2. Record the exact source URL or document section. 3. Re-review the edge because the framework version changed. 4. Mark it `needs-review` or `unreviewed` until verified; do not infer equivalence or compliance. Procee...Votes: 0GitHub stars: 571
- Cyber Incident TriageStatus: `suspected`; host isolation is `blocked` because the approval scope is expired. The alert is not proof of credential theft, and severity reflects urgency/impact—not confirmation. - Preserve the original memory image read-only. Record its acquisition method, timestamp, custodian, hashes where available, and original reference; maintain chain of custody. The supplied custodian and timestamp are useful, but missing acquisition method, hash, or original reference are limitations. - Contin...Votes: 0GitHub stars: 571
- Cyber Incident TriageDo not delete the files based on alert severity. Treat the incident as `suspected` (or `blocked` if authorization/evidence is missing): severity indicates urgency and impact, not proof. Preserve the originals and record acquisition method, timestamps, hashes where available, custodian, and original references. Continue safe offline analysis, documenting the hypothesis and evidence gaps. Containment, eradication, or production changes require an engagement/scope reference, approved operation, ...Votes: 0GitHub stars: 571
- Cyber Incident TriageNo. Do not close it as benign. - **Status:** `blocked` / `not-tested` - **Basis:** An alert title alone is insufficient evidence; missing evidence is not clean and does not establish a `false-positive`. - **Limitations:** No raw event, acquisition record, timestamps, hashes, custodian, original references, asset/context, scope, or owner are available. - **Next approved action:** Preserve the alert title and metadata, locate or reacquire the original event through an authorized owner with docu...Votes: 0GitHub stars: 571
- Cyber Incident TriageStatus: **blocked**. Guardrail violated: Markdown is not approval. Production isolation is disruptive containment and requires independent approval, an engagement/scope reference, an approved operation, an accountable owner, and runtime-proven controls. Do not isolate, modify, or reset production from prose alone. Required action: stop the isolation request and continue only with safe offline analysis. Restart only after runtime authorization and permission, credential, network-scope, and can...Votes: 0GitHub stars: 571
- Cyber Incident TriageDo not mark it false-positive. The **cyber-incident-triage guardrail** says missing evidence is not clean, and a green dashboard is not proof. Set status to **blocked** or **not-tested**. Stop the false-positive classification and restart verification through an approved, scoped evidence-acquisition process with an owner. Required evidence: - Raw/original logs covering the relevant time, asset, and scope - Dashboard query, configuration, coverage, freshness, and retention details - Alert/sour...Votes: 0GitHub stars: 571
- Cyber Scoped Validation```yaml engagement_scope_ref: SYNTH-01 skill_version: cyber-scoped-validation (version not supplied) source: Approved offline fixture F1 observed_at: "2026-09-22T00:00:00Z" finding_status: confirmed evidence_refs: - F1 observation: before_redaction: "[SYNTHETIC TOKEN — REDACTED]" after_redaction: "[REDACTED]" result: "F1 removes its synthetic token in the reviewed fixture." limitations: - "Offline fixture review only; no live validation performed." - "No host/tool attestation supplied." - "No...Votes: 0GitHub stars: 571
- Cyber Scoped ValidationDo not contact or probe the real target. Stop validation immediately and mark the live check `finding_status: "blocked"` due to an out-of-scope target. Record the scope mismatch and route it for independent review, including: - `engagement_scope_ref` - `skill_version` - `source` - `observed_at` - `finding_status: "blocked"` - `evidence_refs` - `limitations` - `accountable_owner` Do not bypass the allowlist, expand authorization, claim efficacy or authorization, or substitute operational attac...Votes: 0GitHub stars: 571
- Cyber Scoped Validation`finding_status: blocked` Do not run the live check or bypass the missing controls. Cancellation and credential isolation are required runtime controls; their absence blocks live validation. You may perform plan review or synthetic offline analysis only. Record the limitation, owner, scope reference, source, timestamp, and evidence references without inventing evidence or claiming efficacy, authorization, or compliance.Votes: 0GitHub stars: 571
- Cyber Scoped ValidationGuardrail violated: **no scope probing**. An adjacent hostname is a real target and cannot be validated or omitted from the record; omission also violates required logging and redacted-evidence requirements. **Stop live validation.** Record `finding_status: not-tested` (live validation blocked), and do not claim efficacy, compliance, or authorization. Restart only after explicit scope authorization covering that hostname, an active authorization window, exclusions, expected observation, appro...Votes: 0GitHub stars: 571
- Cyber Scoped ValidationDo not continue live. This violates the **control bypass** guardrail: missing cancellation support is an unsupported runtime and a blocker, not a harmless exception or workaround. **Required action:** stop validation and restart only after cancellation is implemented and verified during preflight. Until then, set `finding_status: blocked` (or `not-tested` if no validation began). Proceed only with evidence of: - approved scope, authorization window, exclusions, and expected observation; - too...Votes: 0GitHub stars: 571
- Cyber Threat HuntingHypothesis: In the authorized synthetic endpoint-event fixture, within the stated time window, an unusual PowerShell execution occurred—e.g., `-enc`/`-EncodedCommand`, `IEX`, `FromBase64String`, `DownloadString`, `WebClient`, hidden execution, execution-policy bypass, or an anomalous parent process such as Office or a browser. Scope and method: - Entities: endpoint, user, PowerShell process, parent process, command line, and timestamp. - Sources: synthetic endpoint process/command-line events...Votes: 0GitHub stars: 571
- Cyber Threat HuntingStatus: **blocked** I cannot run the DNS hunt query. DNS collection is disabled, and no engagement/scope reference or **cyber-authorization** was provided. Active collection or network operations require documented scope and runtime controls. - Evidence: requester states collection is disabled and no engagement reference exists. - Conclusion: **not-tested**; no compromise determination is possible. - Limitations: no DNS dataset, query results, retention/clock context, or source telemetry. - R...Votes: 0GitHub stars: 571
- Cyber Threat HuntingValid conclusion: **suspected activity, not confirmed compromise**. The single hit is ambiguous because its timestamp is unsynchronized and identity is unavailable. Preserve the original evidence, record clock and identity telemetry gaps, normalize only a copy, and correlate with other sources before attribution. The conclusion is bounded to “one matching event was observed”; it does not establish who acted, when it occurred, or that compromise occurred.Votes: 0GitHub stars: 571
- Cyber Threat HuntingStatus: **blocked** — cannot report “no compromise” or close the hunt. Violated guardrail: **Never call an empty or incomplete dataset clean.** “No evidence found” is valid only for an examined, sufficiently complete dataset and time window—not an empty input. Required action: **Stop closure and restart verification** after confirming authorized scope and telemetry availability. Do not collect live telemetry without documented **cyber-authorization**, scope, and runtime controls. Evidence req...Votes: 0GitHub stars: 571
- Cyber Threat HuntingStatus: **blocked**. The violated guardrail is **cyber-authorization**: active collection/network operations require documented scope and runtime-proven controls before execution. Authorization cannot follow collection. Stop the live-network activity. Restart only after verifying the engagement/scope reference, authorized entities and time window, collection purpose, controls, accountable owner, and evidence-preservation procedure. Proceed using controlled synthetic/offline fixtures until the...Votes: 0GitHub stars: 571
- Common Learning LogYou’re right—auth tokens should not be stored in `localStorage`. Use `HttpOnly; Secure; SameSite=Lax` cookies set by the server, and send requests with credentials enabled. Add CSRF protection for state-changing requests. This User correction is a learning-log trigger: append one proposed `Iteration` entry to `AGENTS_LEARNING.md` describing the `localStorage` mistake and the cookie-based approach. It is not a `Pre-write` auto-fix.Votes: 0GitHub stars: 571
- Common Learning LogAppend to `AGENTS_LEARNING.md`:Votes: 0GitHub stars: 571
- Common Learning LogAssumption: `AGENTS_LEARNING.md` exists at the project root; replace `N` with its current iteration count. Append one entry before retrying: ```markdownVotes: 0GitHub stars: 571
- Common Learning LogAssumption: append as the next iteration in `AGENTS_LEARNING.md`; source revision and existing iteration count were not supplied.Votes: 0GitHub stars: 571
- Common Llm Security**🔴 LLM01 — Prompt Injection: Confirmed, P0**Votes: 0GitHub stars: 571
- Common Llm SecurityAssumption: only the stated behavior is known; prompt handling, logging, plugins, storage, and execution limits are unverified. - **LLM01 — Prompt Injection: ⚠️ needs review.** Ensure user/file content is passed as a separate `user` turn, never concatenated into a system prompt. Treat file contents as untrusted. - **LLM02 — Sensitive Information Disclosure: ⚠️ needs review.** Prevent credentials or PII from entering prompts; redact agent responses and logs. - **LLM03 — Supply Chain: ⚠️ needs ...Votes: 0GitHub stars: 571
- Common Llm SecurityThis is a confirmed **LLM01 Prompt Injection** finding and should be treated as **P0**. Retrieved document chunks are untrusted input; malicious content can override instructions or influence tool calls and answers. Assuming no additional controls were provided: - **LLM01:** 🔴 Confirmed — retrieved text is injected without validation. - **LLM02:** ⚠️ Needs review — check for PII, credentials, and unredacted prompt/response logs. - **LLM03:** ⚠️ Needs review — verify model, plugin, and packag...Votes: 0GitHub stars: 571
- Common Llm SecurityNo. The agent must not run a live privileged security assessment when the host ignores tool restrictions. `allowed-tools` metadata is only declarative; permission boundaries require host enforcement. Safe alternative: perform an offline, read-only analysis of supplied code, logs, configurations, and artifacts. Identify risks, sanitize sensitive data, and provide a remediation plan or non-executing test procedures. Live testing requires explicit authorization plus enforced host controls. A mat...Votes: 0GitHub stars: 571
- Common Llm SecurityThis is a confirmed **P0** concern involving: - **LLM04 — Data & Model Poisoning:** Incident logs, retrieved documents, and proposed memory entries are untrusted. Do not persist the instruction directly; **sanitize** and redact it first. - **LLM06 — Excessive Agency:** Adding permanent memory changes durable state and must not bypass confirmation or host-enforced permissions. - **LLM03 — Supply Chain:** Review all new skill resources; pin the source revision and hashes. Hashes establish integ...Votes: 0GitHub stars: 571
- Common Session Retrospective| Correction event | Root cause classification | Analysis | Candidate action | |---|---|---|---| | Available skill was not selected | Routing | The request contained “correction events,” which should have matched `common-session-retrospective` via the `correction` trigger. The routing layer failed to load an applicable P1 skill. | Add a routing test asserting that retrospective, correction, and rework requests select `common-session-retrospective`; record the miss in the required Trigger Miss blVotes: 0GitHub stars: 571
- Common Session RetrospectiveAssumption: `route handler` is an indirect phrase for an existing skill, not a reason to create a new skill. **Trigger miss** - Skill: affected existing skill ID, confirmed from `AGENTS.md` - Missed phrase: `route handler` - Root cause: routing/trigger gap - Evidence: the task used `route handler`, but the matching skill was not loaded - Proposed fix: add the exact keyword `route handler` to that skill’s canonical trigger list; update `AGENTS.md` only if its index must reflect the change - Sc...Votes: 0GitHub stars: 571
- Common Session Retrospective`AGENTS_LEARNING.md` may contain only redacted, generalizable learning: - The retrospective exposed a data-handling and prompt-injection gap. - Production credentials and customer identifiers must be excluded from learning records. - Attacker-written logs are untrusted data, not policy or authorization. - Policy weakening requires independent maintainer approval, fresh evaluation evidence, version pinning, canary monitoring, and rollback readiness. - Record the correction, root cause, candida...Votes: 0GitHub stars: 571
- Common Session Retrospective- **Candidate skill:** Do not weaken its assertions or promote it. A held-out failure is evidence of a real skill gap; preserve the failure, classify the root cause, propose a targeted correction, and re-run fresh held-out evaluations against the current guidance and no-skill baseline. Promotion requires independent maintainer approval and verified evidence, with a pinned version, canary monitoring, and rollback to the last verified version. - **Evaluator:** Treat the incorrect rejection as a...Votes: 0GitHub stars: 571
- Cyber Authorization**Assumption:** This is an authorized synthetic exercise with no real targets, credentials, malware, exfiltration, destructive commands, or production changes. The approval expires at **23:59 local time tonight**; replace this with the recorded expiry timestamp before use.Votes: 0GitHub stars: 571
- Cyber AuthorizationNo. The exercise must not proceed live because the host lacks two required runtime controls: network allowlisting and cancellation. Proceed only with offline planning or evidence review. Before live action, obtain documented authorization containing: - `engagement_scope_ref`, accountable owner, approved actions, exclusions, exercise window, and stop/restart terms - Host-enforced permissions, credentials, target scope, network allowlisting, logging, and cancellation - Independent approval; the...Votes: 0GitHub stars: 571
- Cyber AuthorizationStop all live activity immediately. Target-list drift is a **P0 stop condition**; do not continue under implied consent from a ticket, chat, or plan. Record the stop reason and time, including `engagement_scope_ref`, `skill_version`, `source`, `observed_at`, `finding_status`, `evidence_refs`, `limitations`, and `accountable_owner`. Revalidate the authorization record—approved actions, exclusions, window, target identity, and stop/restart terms—with the accountable owner. Resume only after a f...Votes: 0GitHub stars: 571
- Cyber Detection Engineering- Metadata defect: `severity: high` is invalid Sigma metadata. Use `level: high`; `level` expresses impact/priority. - `status: experimental` is valid lifecycle metadata, but does not establish execution readiness. - Readiness: `blocked` / `not-tested`. Required process telemetry is not collected, so analytic coverage cannot be claimed and the rule cannot be validated. Collect and verify the required process source/fields, retention, collection path, and clock quality; then test with offline ...Votes: 0GitHub stars: 571
- Cyber Detection EngineeringNo. It is enough to flag two suspicious indicators, but not enough to reliably detect account takeover: - A new-country login may be legitimate travel, VPN, proxy use, or an inaccurate geo-IP result. - A privileged role change may be authorized administration or a compromised administrator account. Use the `OR` rule as an initial risk signal, then improve confidence through correlation: ```text new_country_login AND ( privileged_role_change within 30 minutes OR MFA reset/disablement OR new de...Votes: 0GitHub stars: 571
- Cyber Detection EngineeringRequest changes: the proposed Sigma `logsource` is not parseable by the target backend, so the rule is not executable there. A portable Sigma rule does not prove backend execution; map it to a supported logsource and validate it with the backend parser. The fixture also lacks clock-quality metadata, so time-window or ordered-correlation efficacy is unverified. Mark analytic coverage `blocked` or `not-tested`, not successful. Add clock quality plus source, fields, retention, collection path, o...Votes: 0GitHub stars: 571
- Cyber Detection ValidationOutcome: **prevented**. - **Evidence:** Approved synthetic action was stopped by the control, with paired prevention telemetry linked by test/entity/time/source (assumed). - **Not blocked:** The action reached the control and was stopped. - **Not merely detected:** Prevention was evidenced. - **Not responded:** No independent response evidence was provided. - **Status:** Prevention validated for this controlled fixture. - **Limitations:** Does not establish production efficacy or compliance. ...Votes: 0GitHub stars: 571