Agent-Skills.md

Agent Skills: dogfood-ledger

Count real dogfood cycles per agentic-tool (the ≥2-cycle promotion gate authority). Use to mark a dogfood cycle (in-progress / complete+ratified+evidence) or to tally how many real cycles a tool has before promotion. Replaces changelog-prose cycle counting with a structured, jq-countable, auditable ledger. Triggers - "how many dogfood cycles does X have", "mark this cycle complete", "is X promotion-eligible", "count cycles".

UncategorizedID: ekson73/multi-agent-os/dogfood-ledger

Repository

ekson73/multi-agent-os
ekson73License: MIT

Install this agent skill to your local

pnpm dlx add-skill https://github.com/ekson73/multi-agent-os/tree/HEAD/skills/dogfood-ledger

Skill Files

Browse the full folder contents for dogfood-ledger.

Download Skill

Loading file tree…

skills/dogfood-ledger/SKILL.md

Skill Metadata

Name
dogfood-ledger
Description
Count real dogfood cycles per agentic-tool (the ≥2-cycle promotion gate authority). Use to mark a dogfood cycle (in-progress / complete+ratified+evidence) or to tally how many real cycles a tool has before promotion. Replaces changelog-prose cycle counting with a structured, jq-countable, auditable ledger. Triggers - "how many dogfood cycles does X have", "mark this cycle complete", "is X promotion-eligible", "count cycles".

dogfood-ledger

Thin ergonomic wrapper over the AAIF bash CLIs bin/dogfood-mark + bin/dogfood-tally. The engine is vendor-neutral bash (works in Cursor / Copilot / Codex / Gemini too). Spec: docs/dogfood-cycle-ledger-spec.md · ADR: docs/adrs/ADR-005-dogfood-cycle-ledger.md

Purpose

The dogfooding-mandate gates promotion on ≥2 real cycles, but the count was only changelog prose — unmeasurable = theater. This skill makes it real: one user-scope, append-only, jq-countable ledger at ~/.claude/audit/dogfood-cycles.jsonl, fed by any session/project/vendor. bin/dogfood-mark is the writer; bin/dogfood-tally is the read-only counting authority.

When to Use

Use when you need to record or count dogfood cycles for the ≥2-cycle promotion gate.

Count cycles for a tool (the gate authority):

bin/dogfood-tally debate-converge          # table: COMPLETE / IN-PROGRESS / ABANDONED + gate verdict
bin/dogfood-tally                           # all tools
bin/dogfood-tally debate-converge --json    # machine-readable (+ .eligible)

Mark a cycle:

# open / advance (in-progress)
bin/dogfood-mark debate-converge 001 --status in-progress --evidence session:<sid>

# close it (COMPLETE requires --ratified AND >=1 --evidence — anti-theater)
bin/dogfood-mark debate-converge 001 --status complete --ratified \
  --evidence github:ekson73/multi-agent-os/pull/97 --evidence session:<sid>

Backfill real historical cycles (cycles that already happened, never counted) — Phase 2, planned (not yet implemented):

bin/dogfood-mark --backfill --dry-run       # (planned) show detected cycles + evidence
bin/dogfood-mark --backfill                  # (planned) register them (evidence-based, anti-hallucination)

Until --backfill lands, record historical cycles with direct evidence-based dogfood-mark calls.

Trigger Phrases

  • "how many dogfood cycles does X have"
  • "mark this cycle complete" / "mark cycle in-progress"
  • "is X promotion-eligible" / "did X meet the ≥2-cycle gate"
  • "count cycles" / "tally dogfood cycles"

Protocol Rules

  • A cycle is COMPLETE only with ratified==true + ≥1 evidence ref. "Rounds happened" ≠ complete.
  • The ledger is the single SSOT — do NOT keep per-skill island counters (they migrate here).
  • dogfood-tally is read-only; all writes go through dogfood-mark.
  • Local-only ledger → LGPD long-term retention is an out-of-band gap (do not over-promise permanence).

Composition

  • ASH bin/ash-walkthrough --dogfood-cycles (vkl) reads this same ledger (consumer, not a 2nd SSOT).
  • CPT §9 Compass API + gsd-graphify consume the link-fields for inter-relation/graph (this skill does not visualize).