Content
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is highly actionable with a clear, well-validated workflow for a risky host-hardening task, but it is somewhat redundant across sections and monolithic where progressive disclosure into reference files would help. Strengths are concrete commands and explicit checkpoints; weaknesses are token efficiency and file structure.
Suggestions
Dedupe repeated guidance — the model self-check, the 'Bitterbot does not change host firewall/SSH/OS updates' caveat, and the confirmation lists each appear in multiple places; state each once and cross-reference.
Move large inline reference material (the cron-scheduling procedure, memory-write format, full context-gathering prompt bank) into separate reference files under ./references/ and link to them one level deep, keeping SKILL.md a lean overview.
Tighten verbose phrasing in Core rules and the periodic-checks section to assume Claude's competence and cut tokens that restate workflow steps.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean and directive, but several points are repeated across sections (model self-check in Core rules and Section 0; 'never claim Bitterbot changes firewall/SSH' at lines 18, 88, 214; overlapping confirmation lists), so it could be tightened. | 2 / 3 |
Actionability | Provides exact, copy-paste-ready commands (uname -a, ss -ltnup, ufw status, bitterbot security audit --deep [--fix] [--json], bitterbot cron add|list|runs|run), stable cron names, numbered execution options, and a strict CLI flag allowlist — fully concrete guidance for an instruction-only skill. | 3 / 3 |
Workflow Clarity | A clear 0–8 ordered workflow with read-only context gathering first, explicit validation re-checks in Section 8, rollback/access-preservation planning, and a 'stop on unexpected output' feedback loop for destructive firewall/SSH/port operations. | 3 / 3 |
Progressive Disclosure | Well-organized into clear sections, but as a ~240-line single file it inlines substantial detail (context-gathering prompts, full cron-scheduling procedure, memory-write format) that could be split into one-level-deep reference files rather than kept monolithic. | 2 / 3 |
Total | 10 / 12 Passed |