Compaction survival
Skill markoblogo/abvx-agent-skills/skills/compaction-survival
Reviewable capability layer for coding agents. Portable skills, delivery gates, workflow patterns, and verification-first engineering.
npx -y skills add markoblogo/abvx-agent-skills --skill compaction-survivalAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Preserve working state across long sessions and context compaction by checkpointing decisions, edited files, blockers, open hypotheses, and unresolved errors into compact resumable artifacts. Use on long coding tasks, audits, research, or any session at risk of losing state after compaction.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
2.1 KB, as published. Nobody here has run it
Compaction Survival
Save tokens by keeping the right state alive across compaction instead of re-deriving it after memory loss.
Goal
When a session is getting long, preserve only the working state that would be expensive to reconstruct.
What To Preserve
- current objective;
- active subtask list;
- key decisions and reversals;
- files changed or inspected deeply;
- unresolved errors and their current hypothesis;
- blockers, constraints, and pending verifications;
- exact artifact paths for large outputs or generated files.
What Not To Preserve
- long narrative history;
- repeated summaries of already-resolved steps;
- raw logs that can be regenerated cheaply;
- generic project documentation already available on disk.
Checkpoint Workflow
- Detect compaction risk:
- long session;
- many files touched;
- many turns of debugging or research;
- user explicitly worried about token budget.
- Emit a compact handoff artifact:
- objective;
- done / next / blocked;
- decisive evidence only;
- exact paths for deeper artifacts.
- Keep the checkpoint stable unless the working state materially changes.
- After compaction or resume, recover from the checkpoint first instead of replaying the whole session.
Recommended Shapes
- concise handoff note;
- checked-location ledger;
- issue-specific status file in
docs/ai/tasks/or equivalent; - compact bullets over narrative prose.
Guardrails
- Do not checkpoint secrets.
- Do not create noisy per-turn diary files.
- Do not preserve stale hypotheses after the evidence changed.
- Do not replace real verification with checkpoint confidence.
Final Report
Say what was checkpointed, where it lives, and what should be reused first after resuming.