cellm/skills/mantras/SKILL.md
Four CELLM work process mantras as mental gates — verify, best path, document, resolve. Active by default for state-changing actions with non-trivial scope or irreversible side-effects. Trivial actions pass silently.
npx skillsauth add murillodutt/cellm cellm/skills/mantrasInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
From Sanskrit: man (mind) + tra (instrument/protection). Literally: "instrument that protects the mind."
The mantra is a mental gate that interrupts shortcuts before they cause harm. The gate's purpose is judgment, not ceremony.
"Verify before you act, take the best path — never the first, document everything, and resolve what you find — never normalize debt."
| Action class | Gate active | Notes | |--------------|-------------|-------| | Irreversible side-effect (push, delete, send, pay) | Yes, explicit | All 4 gates must be satisfied before action | | Multi-file change or schema/contract edit | Yes, explicit | Document gate must record the spec/commit reference | | Single-file edit with clear local intent | Yes, silent | Internal check; no output unless a gate fails | | Trivial action (rename variable, fix typo, format) | Silent | The gate exists in mind, not in output | | Investigation or read-only audit | Silent | No state change → no ceremony |
The gate is judgment, not output. Silent passage is valid when all four criteria are obviously satisfied. Verbose output is required only when at least one gate is borderline or when the action is in the irreversible class.
| Gate | Challenge | Pass condition | Fail signal | |------|-----------|----------------|-------------| | VERIFY | "What evidence do I have?" | Concrete: grep result, file read, DB query, doc read, schema | "I think...", "should be...", "the format is..." | | BEST PATH | "Is this the best approach or just the first?" | If 2+ paths exist, evaluate; if only 1 obvious path exists, take it. Do not fabricate alternatives. | Defaulting to the first path when a clearly better one was visible | | DOCUMENT | "Where is this recorded?" | Spec node, commit message, knowledge atom, or journal entry exists or will exist in this turn | Action taken without any persistent trace | | RESOLVE | "Did I fix what I found?" | Problems identified are addressed in the same cycle, OR explicitly deferred with a tracked spec/issue | "Not my problem", "we'll fix later" |
When the action class requires explicit gate output, one line per gate is enough:
[MANTRA GATE]
VERIFY: {what was checked, source}
BEST PATH: {why this path; alternatives if any}
DOCUMENT: {where this is recorded}
RESOLVE: {problems addressed}
→ PROCEED
For trivial actions, this output is omitted entirely. The gates ran in the mind; the action proceeds.
In Olympus mode, gate violations are findings.
| Metric | What it measures |
|--------|-----------------|
| verify_failures | Times evidence was assumed, not checked |
| bestpath_failures | Times a clearly better path was visible but the first one was taken |
| document_failures | Actions completed with no trace in any system |
| resolve_failures | Problems identified but normalized instead of fixed |
Note: gates_skipped was removed. Skipping silent gates is the design — counting it as failure rewarded ceremony over judgment.
data-ai
Prose override — temporarily disable quantization and respond in readable prose. Use when relational density matters, for safety-critical explanations, onboarding handoffs, or when token economy is not the priority.
development
Govern explicit weekly Super PRs or maintainer-requested PR merges. Evaluates a 10-criterion readiness checklist and performs governed merge only when a user-requested PR is READY. Never creates or keeps permanent PRs. Use when: 'pr-check', 'pr-merge', 'merge this PR safely', 'is PR ready', 'guard merge', or /sk-git delegates pr-merge.
data-ai
Operational surface for the compress-llm Layer-1 token I/O compressor. Enable, disable, switch mode, and inspect status without editing config files. Use when tuning compression pressure for the current session or project.
tools
Generate structured upstream feedback for the CELLM engineering team. Produces evidence-first Markdown at docs/evidence/<date>-cellm-feedback-*.md for bugs, anti-patterns, deprecation gaps, and harness surprises, with optional atom registration via knowledge_ops. Use when: 'feedback for CELLM', 'send to CELLM team', 'register this as atom', 'document this anti-pattern', 'report this bug upstream'. Trigger proactively on MCP schema/runtime mismatches, mechanical edit loops (>=3 sequential edits), short deprecation windows (<6 weeks), or reusable harness surprises. Do NOT trigger for routine feature work or project-local bugs.