Lintel on GitHub Copilot. Generated from
skills/sc/SKILL.md; edit the canonical file, then runli-copilot init.
- Resource root:
../../..from this skill's base directory (the Lintel source withbin/,lib/,skills/). Write plans, state and evidence into the working repository's.claude/tree, never into the resource root.- Skill-relative paths: this skill's own
scripts/,references/anddata/folders (and a<base>that the workflow defines as its own directory) mean../../../skills/sc/in the Lintel source, not this generated folder.${LINTEL_SKILLS_DIR:-skills}means the skills root,../../../skills. Abin/li-runstep runs in the working repository, so use$LINTEL_SKILLS_DIR/sc/there.- Shell steps: run Bash snippets with Bash (Git for Windows'
bash.exeon Windows, neverSystem32\bash.exe). Save a snippet to a temporary.shfile and runbash "<resource root>/bin/li-run" <file>; it preparesLINTEL_SOURCE_ROOT,LINTEL_REPO_ROOTand the profile context.- Tools: Read=
view, Write=create, Edit=edit, Bash=bash/powershell, Grep=grep, Glob=glob, AskUserQuestion=ask_user, TodoWrite=the plan checklist, Task or a named role=taskwith that custom agent, WebFetch=web_fetch.- Other Lintel workflows are native skills: invoke
/li-<name>rather than reading their files. Named roles such asCodeReviewerare custom agents.
Security and compliance
Preference metadata only: optional engineering preferences
explains the retained preferences_root hint; it is not a validated pack interface.
Read security decision methods: implementation evidence, legal/policy applicability and unknowns are different. Produce threats, auth/secret/audit designs, compliance coverage and incident playbooks, not a blanket certification. The host's permission and actual policy remain authoritative.
For an implementation review, use the same adaptive Review Method and targeted security evidence as REVIEW and MARS. Keep source/version, actual trust boundary and selected mandatory controls explicit. Object/tenant authorization is distinct from privacy or residency; agent/tool authority is distinct from a model's declared role. This adds review questions, not a scanning engine, credential use or permission to reproduce vulnerabilities.
/li-sc full covers the checkpoints below. /li-sc loop revisits explicitly saved evidence;
/li-sc <capability> or /li-sc single --action <capability> limits the assignment. Preserve
read-only review versus authorized implementation. No scan of live services,
customer data, credential use/rotation or incident message follows from invocation.
Sub-capability dispatch
| Capability | Dispatches to (agents) | Produces | Decisions and checks |
|---|---|---|---|
threat-model |
ThreatModelDrafter, then distinct SecurityAuditor review | STRIDE by boundary, mitigations and residual-risk decisions | actual assets/actors; platform-covered threats need deployment evidence, not omission |
secret-management |
SecurityAuditor + SecretsScanReviewer; SBOMAuditor for dependency-shipped defaults | redacted inventory, owner, storage, response/rotation and default-credential triage | no values or live credential tests; cadence/provider comes from applicable risk/policy |
auth-flow |
OAuthFlowReviewer for OAuth/OIDC; JWTSecurityReviewer for JWT internals; SecurityAuditor for other auth | flow diagrams, token/claim/scopes/revocation checks | one protocol is not another; grant/client-specific checks, exact library/provider versions |
compliance-evidence |
ComplianceOfficer per applicable framework; Architect for system boundary | technical/procedural evidence, cross-framework reuse and gap map | primary/consolidated source/version, effective/application dates, verified_on, responsible owner and currency limits; no default SOC2/GDPR or score-based certification |
audit-path |
Architect design + SecurityAuditor review | event/integrity/transport/retention design and checks | actor/action/target/outcome/context; UTC timestamp and ordering mechanism distinct; no blanket seven years |
dependency-security |
DependencyAuditor + SBOMAuditor | ecosystem SCA and artifact-scoped SPDX/CycloneDX evidence | actual advisory range/reachability and license use/distribution policy; unavailable feed remains unverified |
incident-runbook |
SecurityAuditor + DeploymentEngineer | detection, containment, eradication, recovery, escalation and postmortem | selected threat/observability evidence, named on-call/live-action authority; no real containment or notifications |
For auth, retain login/refresh/revoke/step-up diagrams, justified lifetimes, refresh/ revocation paths, role-specific MFA and replay/rate-limit/session-fixation checks. For audit, consider append-only/WORM or tamper evidence only when justified; prove transport loss handling and queryability. Policy sources, not templates, set retention.
Workflow
Read and execute the shared module caller procedure before domain work. It is the sole owner of original work/package/leaf admission, live P07/policy checks, immutable obligations, checkpoint publication, cold continuation and final QA/independent acceptance. Apply this security method inside that procedure:
- Read the selected requirements and actual threat/data boundaries. Obtain typed verified pack fields or explicit advice; no framework is mandatory merely because the module knows it. Unresolved applicable policy stops its affected action.
- Use the capability and checkpoint tables to distinguish design, implementation evidence and legal/control applicability. Keep actual review versus planning mode; only authorized synthetic/local observations are evidence, not suggested scans.
- Retain redacted findings, missing observations and the residual-risk owner's decision. A reviewer reports and does not fix or silently accept the risk.
Checkpoint ownership
| Checkpoint | Method and owner | Observable acceptance |
|---|---|---|
threat_model_complete |
ThreatModelDrafter | trust boundaries, concrete failure paths, mitigation owners and explicit residual decisions |
secrets_inventoried |
SecretsScanReviewer/SecurityAuditor | redacted location/type/owner/scope, actual configured response and missing evidence |
auth_flow_locked |
Applicable flow/token reviewers | correct client/grant/profile checks with implemented-versus-planned evidence |
compliance_evidence_present |
ComplianceOfficer | every required applicable control has source/version/evidence/outcome; grounded N/A is distinct from unknown |
audit_path_verified |
SecurityAuditor | required-event producer-to-sink observation and retention/integrity evidence, or explicitly unverified |
Worked contrast: signature validation on a correctly signed ID token does not establish API authorization. Trace issuer/audience/type to the decision, then test the wrong-audience case in an authorized synthetic fixture. A missing SDK setting is not proof of a leak; a policy brochure is not proof of configured protection.
Handoff, scoring and recovery
Preserve named threat model, secret-management plan/inventory, auth-flow review,
per-framework coverage/reuse/gap reports, audit-path and incident runbook artifacts.
Bind them to the selected request under safe attempt paths; old state/sc/ outputs
are selected evidence only, never the newest-file default.
The six rubric dimensions remain Threat model coverage, Mitigations declared, Secrets inventoried, Auth flow review verdict, Compliance evidence and Audit path verified, each measured against the observable checkpoint criteria. Scores are advisory; one mandatory unresolved result blocks regardless of five high scores. DONE requires required evidence and independent acceptance. DONE_WITH_CONCERNS has advisory residuals only; missing policy/authority/evidence is BLOCKED or NEEDS_CONTEXT.
Use the shared cold-attempt table. Missing result after start is interrupted; do not rerun scans, notify owners or rotate keys on inferred permission. Loop with explicit new input identity and prior evidence links. DeploymentEngineer provides planning-only recovery/on-call content alongside SecurityAuditor's security response reasoning; the caller persists it. ReleaseEngineer retains authorized release execution for a separately authorized invocation, not this planning dispatch.
Reusable patterns
Follow the reusable pattern consumer contract.
Module entry resolves the security expectations bound to the current target context and passes
each sub-capability only its projected clauses (project for the mapped package, or the clauses
mapped to it). An unknown deployment target or trust boundary yields needs-context and blocks
the dependent design; no live cloud, tenant or environment discovery is implied, and
requirements are never inferred from generic recommendations. When the runtime reports no
patterns, the module is unchanged.
Integration and dormant hooks
TA/DA supply system/data boundaries, DH consumes response/audit needs, TQ tests the actual mitigations. No new policy parser, scheduler, phase or release authority. Optional Brief Forge/audit invocation is explicit, with observed persistence. No active control is inferred from these opt-in domain files (ADR-0008):
hooks/shared/sc-threat-coverage-warn/hooks/shared/sc-auth-bypass-warn/hooks/shared/sc-compliance-gap-warn/
Existing secret/customer-data hooks retain their host-specific registration and permission boundaries. Optional domain warnings remain unobserved without actual execution evidence and cannot replace the required review or control outcome. Never claim a fictional no-production-mutation hook fired.