Back to skill

Security audit

telo: Apple Health

Security checks for vulnerabilities and agentic risk

Overview

The scanners report a benign skill with no AIG findings; Skillspector flagged retry wording as a possible resource-use concern, but its review disagreed with that finding.

The verdict is benign with medium confidence, but benign does not mean risk-free. Consider that Skillspector’s analysis was partial: it could not resolve a local path-like reference in SKILL.md, even though all three listed components were scanned. The report does not explain the install-mechanism note, so the available evidence does not fully clarify that dimension.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · references/long-term-trends.md (reported line 24)May include surrounding context.

md
On `truncated: true`, discard the incomplete response before combining results. First request fewer metrics, then split into non-overlapping ranges at full bucket boundaries (month starts for monthly data, Mondays and January 1 for the existing weekly buckets). Keep the original outer bounds. Retry only the affected batches.

Prefer not to split a bucket. If it must be split, combine sums by addition, means weighted by `days_present`, minima/maxima with min/max, and coverage only across disjoint subranges. Never average averages equally or sum repeated buckets from overlapping retries. If a single bucket and metric still truncates, report that the full result could not be retrieved; do not loop indefinitely.

For a suspicious change, query daily values around that period. Use workout sessions or details only if needed to answer the user's question. Follow `workout_detail`'s independent cursors; historical summaries may exist after raw series expire.

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · references/weekly-health-review.md (reported line 27)May include surrounding context.

md
mplete totals as evidence that activity fell. Do not extrapolate missing days.
5. If available and relevant, call `derived_metrics` for the recent window to contextualize HRV/resting-HR against preceding personal baselines. Respect `insufficient_data` and the reported baseline sample count. Training load is duration in minutes, not intensity-weighted TSS or a readiness score; its zero-workout days may include missing uploads.

If a response is truncated, discard it as an incomplete result and retrieve smaller non-overlapping date ranges or metric batches. For daily queries, splitting dates does not split daily buckets. Deduplicate overlapping retries by date/metric/unit/aggregation and sessions by UUID. If even a single day for a single metric or session kind is truncated, disclose the unresolved limit; do not fabricate a complete total or retry indefinitely. Use `workout_detail` only for a requested drilldown and follow its independent series and event cursors.

## Present the review