handoff-249215 — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited handoff-249215 (Agent Skill) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Create a structured markdown document that enables a fresh Claude conversation to continue the current task seamlessly. The handoff captures working state — goals, progress, decisions, context, and next steps — so the receiving session picks up exactly where this one left off.
Silently assess:
If the user included specific instructions in their handoff request (e.g. "create a handoff, make sure to include X"), incorporate those.
Skip this step entirely if the session state is clear. Ask at most 2-3 short, specific questions only when a genuine ambiguity would produce a materially different handoff:
Do not ask: open-ended questions ("anything specific you want me to capture?"), questions answered or implied in the conversation, or questions about minor details that would not change the handoff's usefulness.
Build the document using the adaptive structure matching the work type and complexity tier.
Save as a markdown file and present it for download.
Filename: YYYY-MM-DD-handoff-[brief-description].md Use today's date and a short descriptive slug (e.g. 2026-06-11-handoff-epq-silicon-anode-research.md). The date prefix keeps multiple handoffs sortable and prevents collisions.
Classify the session to determine which structural sections are most relevant. If a session spans multiple types, use the primary type and pull in sections from secondary types.
Exploring a topic, gathering sources, evaluating information, forming conclusions. Key sections: Objective, Findings So Far, Sources and Evidence, Open Questions, Methodology, Conclusions Reached, Gaps Remaining.
Producing or editing text: essays, applications, emails, articles, documentation. Key sections: Objective, Current Draft State, Tone and Style Decisions, Structural Outline, Content Completed vs Remaining, Editorial Decisions.
Designing plans, making decisions, evaluating options, setting direction. Key sections: Objective, Options Considered, Decisions Made (with rationale), Constraints, Current Plan State, Unresolved Questions, Dependencies.
Creating something: websites, apps, designs, configurations, data work. Key sections: Objective, Architecture/Structure Decisions, What Has Been Built, Technical Choices (with rationale), File and Asset References, Issues Encountered, Remaining Implementation.
Understanding a concept, working through problems, studying material. Key sections: Objective, Concepts Covered, Key Takeaways, Areas of Confusion Remaining, Resources Used, Topics Still to Cover.
Diagnosing an issue, troubleshooting, iterating toward a solution. Key sections: Objective, Problem Description, Hypotheses Tested, What Worked / What Did Not, Current Diagnosis, Remaining Approaches to Try.
Use the universal structure and pull in type-specific sections as needed.
Size the handoff to the complexity of the work. Be complete on decisions, rationale, and failed approaches; ruthless on everything else. The receiving session pays to read every word, and an overlong handoff buries the next-step signal. The handoff fails only if the user has to re-explain something — not if it omits things Claude already knows.
These counts are calibration guidelines, not targets to fill.
Every handoff follows this skeleton. Include or exclude sections based on work type and relevance. Never include empty sections.
# Handoff: [Descriptive Title]
> **Resumption instruction:** Review this document, then confirm your
> understanding in 2–3 sentences before proceeding with the next step.
**Date:** [YYYY-MM-DD]
**Work type:** [Research / Writing / Planning / Building / Learning / Problem-Solving / Mixed]
**Status:** [In Progress / Paused / Nearing Completion / Blocked]
---
## Objective
[What we are trying to accomplish and why. The north star for the receiving
session — enough context to understand motivation and scope, not just the task.]
## Current State
[A precise snapshot of where the work stands right now — not a summary of the
conversation. Not "we discussed the layout" but "the layout is finalised as a
two-column grid with sidebar navigation; the header is built, the footer is not."]
## Progress
[Concrete outputs and milestones accomplished this session. Concise entries,
not narrative.]
## Decisions Made
[Each decision with its rationale. The rationale prevents the new session from
relitigating settled questions or unknowingly reversing a considered choice.]
- **[Decision]** — [Why this was chosen over alternatives]
## Approaches That Did Not Work
[Directions tried and rejected, with reasons. Prevents the most common failure
mode: the new session re-exploring dead ends.]
## Project Context
[Stable background: constraints, preferences, and domain knowledge established
during the session that does not change between sessions. Only include when
there is meaningful context beyond the objective. Do not duplicate information
already in Claude's memory system.]
## Open Questions
- [ ] [Specific question needing resolution — not a vague area]
## Next Steps
[Ordered and immediately actionable. The first item is the very next thing to do.]
1. [First action — specific and concrete]
2. [Second action]
## Key References
[Links, file names, or sources relevant to continuing. Only if meaningful.]Include:
Exclude:
Before outputting, verify:
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.