ritual-full
Skill4 starsChanged yesterdayApache-2.0View on GitHub
The ritual flow (Think, Spec, Plan, Build, Review, Ship, Reflect on a Chronis epic) with every tool on by default: EGM for brainstorm, draft, review and drift; a Semantic ISA section in every design; two independent fresh-context reviewers; cos hires per step; graphify context; claude-guard hooks. Runs a tool preflight first and stops on anything missing. Triggers on: ritual full, full ritual, full flow, plan this feature with every tool, ritual with egm, ritual with semantic isa.
- Installs packages
In this repository25 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- allsource-incident-statusSkill
- allsource-partnershipsSkill
- ritual-fullSkill
- ritualSkill
- semantic-isaSkill
- admin-healthSkill
- allsource-data-accessSkill
- allsource-debugSkill
- allsource-incident-statusSkill
- allsource-mcp-setupSkill
- allsource-onboardSkill
- chronos-data-flow-embeddedSkill
- chronos-data-flowSkill
- chronos-durabilitySkill
- chronos-embedded-durabilitySkill
- chronos-releaseSkill
- fly-statusSkill
- mammoth-memorySkill
- ritualSkill
- semantic-isaSkill
- seo-aeo-auditSkill
- allsource-customerSkill
- allsource-prime-memorySkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 17
Run each check, print one table, then stop if any row is missing:
- Very specific: line 48
Every design gets a ## Semantic ISA section, written with semantic-isa's DESIGN mode (SKILL.md, template) and scaled to the change:
- Distinctive: line 52
Side-effect inventory: every file write, network call, API, message, deploy and data write the change adds or touches.
- Distinctive: line 86
VERIFY changed, goes in the report's Drift section and in the design, in the same change.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
ritual-full
The same flow as ritual: same steps, same files, same ritual-* lines on
the epic, same gates. Read that skill first; this one states only what changes. Every tool below
is on by default. A missing tool blocks the flow until it is installed or the user says
"continue without <tool>"; never skip one silently. Zero to hero, with install and a worked
example: docs/ritual-zero-to-hero.md.
Preflight (first thing, every flow)
Run each check, print one table, then stop if any row is missing:
| Tool | Check | Install when missing |
Chronis cn | cn --help exits 0; .chronis/ in the main checkout | cargo install chronis, then cn init in the repo |
| EGM plugin | claude plugin list shows egm@egm enabled | claude plugin marketplace add google/elephant-goldfish-model then claude plugin install egm@egm, new session |
| semantic-isa | ~/.claude/skills/semantic-isa/SKILL.md exists | cp -R skills/semantic-isa ~/.claude/skills/semantic-isa from this repo |
| cos | cos ritual check --help exits 0 | clone decebal/chief-of-staff-v2, run task setup in it |
| graphify | graphify --help exits 0; graphify-out/ in the repo, else build it | uv tool install graphifyy then graphify install |
| claude-guard | claude-guard --help exits 0; ~/.claude/settings.json registers claude-guard infra-guard and claude-guard bash-hygiene | cargo install --path gates/rust/claude-guard, then skillsync hooks install |
| skills | brainstorming, pr-proof, systematic-debugging under ~/.claude/skills/ | cp -R skills/<name> ~/.claude/skills/<name> |
Report: preflight: cn ok · egm ok · semantic-isa ok · cos ok · graphify ok · claude-guard ok · skills ok,
with missing (<install command>) in place of ok where it applies. EGM skills set
disable-model-invocation: true, so for every /egm: step below you give the user the exact
command and wait for its output file.
What is on, per step
| Step | On by default | Hands over (in addition to ritual's) |
| Think | user runs /egm:egm-brainstorm; graphify query on the modules the idea touches | concepts_brief.md moved to the dated brief path; graph nodes cited |
| Spec | user runs /egm:egm-draft; a Semantic ISA section; review by /egm:egm-review and a hired ritual-goldfish; resolve with /egm:egm-resolve or /egm:egm-review-loop; cos ritual check | design with ISA section; two review files |
| Plan | one bead per ISA sink whose done-when is its VERIFY test; graph read back (graph.md) | plan with the ISA test matrix as trophy rows |
| Build | cos agents spawn --role builder per ready bead, own worktree; claude-guard hooks on | ISA matrix rows as real tests |
| Review | cos agents spawn --role reviewer; sink-to-VERIFY check | code review naming each sink's VERIFY at file:line |
| Ship | pr-proof evidence; one PR, one commit (ship.md) | the evidence report |
| Reflect | user runs /egm:egm-drift-report on the design; new-sink check | report under docs/reports/ |
Spec: the Semantic ISA section
Every design gets a ## Semantic ISA section, written with semantic-isa's DESIGN mode
(SKILL.md, template)
and scaled to the change:
- Side-effect inventory: every file write, network call, API, message, deploy and data write the change adds or touches. One row each, with reversibility.
- Instructions in the five cores, each with a governance property and a typed binding
(closed
input_schemaandoutput_schema). - Taint: sources (user input, fetched content, file reads, model output) and sinks.
- VERIFY before every sink: a stated, deterministic check. "The model judged it safe" is not one.
- Violations as recoverable exceptions: evidence recorded and the recovery path.
- ISA test matrix: per sink, tainted input refused and VERIFY-cleared input admitted.
A change with no side effects says so in one line with the evidence (no file, network, API,
message, deploy or data write: <the diff's scope>). The Goldfish reviewers check this section
like any other; an inventory that misses a side effect visible in the code is a blocking finding.
Spec: two reviewers, one approval
The two reviews run in separate fresh contexts and never see each other's output. Write each as
its own review file (format in spec.md). Approval needs both clean
against the same design digest; point cos ritual approve --review at the later one and name the
other in the design's review history. Resolve findings, then re-run both. The approval is the
user's: cos ritual approve in their terminal, or, when they tell you to approve in the
conversation, the same command with --yes and their instruction quoted in your report
(spec.md). Agents never approve on their own initiative.
Review: sink-to-VERIFY check
For each sink in the design's ISA table, the reviewer finds the VERIFY in code and its test:
rg -n '<sink call>' lists every call site, and each one is reached only through the named check.
A sink with no VERIFY, or a call site that bypasses it, is a blocking finding. A sink in the code
that the design does not list is drift: back to Spec.
Reflect: new sinks
Diff the shipped code's side effects against the design's inventory. A new sink, or one whose VERIFY changed, goes in the report's Drift section and in the design, in the same change.
Everything else
Finding the step, the ritual-* lines, where the epic and files live, the gates, bugs skipping
Think and Spec, and per-org notes (repo-notes.md) are as in
ritual.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

