Operating workflows for forward-deployed roles running the engines inside a real company's pipeline and data stack. The skills catalog covers building a site once; this tier covers operating a deployed content and growth pipeline continuously. Same authoring contract, extended with role, status, connectors, and autonomy.
New here? Start with GETTING-STARTED.md.
Install the catalog these workflows invoke:
/plugin marketplace add rampstackco/claude-skills
/plugin install rampstack-skills@rampstack
Three tracks run the pipeline, and a cross-track pair spans them:
- Forward-deployed engineer (FDE): stands the pipeline up and keeps its data and deploys honest.
- Forward-deployed marketer (FDM): runs the content and experimentation loop through prove gates.
- Forward-deployed analyst (FDA): owns the numbers, the attribution, and the diagnosis loops.
- Cross-track: Incident Response and Lane Demotion, and Autonomy Review. These operate over every track, one on the event of a regression and one on a periodic cadence.
The ladder is this tier's spine. A workflow's status says what has actually happened to it, not how good the document reads.
- template: designed, not executed as written.
- validated: executed as written, by the engines, on a showcase-designated property, with a PUBLIC linked run record; manual runs the workflow merely describes do not count.
- hardened: validated plus failure modes documented from real incidents; explicitly the home of demotion and post_merge_outcome regression data, so hardened is earned by surviving events, not by prose.
Every shipped file below enters at template. Content Pipeline with Prove Gates is the first to flip: Run 1 executed it end to end on a designated property, so its status is validated with the run record linked, the evidence landing before the claim.
Checks report and fail, engines propose and stop, a human merges; autonomy is earned per lane under the published doctrine, never asserted.
Every workflow carries the same "If a prerequisite is unmet" section, between its prerequisites and its phases. When a phase's required input, tool, access, or data is unavailable or unverifiable, the sanctioned output is a report-blocked statement, and that statement satisfies the phase's done-when; fabricating the missing number never does. The convention exists because instructions should not require what circumstances may not permit, and a done-when that demands numbers with no sanctioned way to stop is an instruction to invent them.
WORKFLOWS.lock is the canonical manifest and the source for any count. The table is the list. Shipped files link, and with this wave every file is shipped, so no row remains defined and publishing.
| Workflow | Track | Status |
|---|---|---|
| Post-Deploy Live Verification | FDE | template |
| Data Surface Integrity | FDE | template |
| Migration with Verification | FDE | template |
| CI Prove-Gate Wiring | FDE | template |
| Warehouse Data Plane Standup | FDE | template |
| Content Pipeline with Prove Gates | FDM | validated |
| Corpus Integrity and Correction | FDM | template |
| Link Graph and Metadata Parity Audit | FDM | template |
| Regulated-Content Compliance Gate | FDM | template |
| Experiment Loop with Pre-Registered Gates | FDM | template |
| Traffic-Drop Triage | FDA | template |
| Conversion-by-Source Diagnosis | FDA | template |
| Revenue Tracking Integrity | FDA | template |
| Incident Response and Lane Demotion | cross | template |
| Autonomy Review | cross | template |
Generated by scripts/build-tested-on.mjs from openaddict.com. CI checks it against the site, so edit the script, not this block.
Run 5 times on each of 3 models. Of the 24 faults planted in 100 published statements, the models found between 64.2% and 75.8%. Pinned at commit a67dd34, run in Claude Code. Full results: openaddict.com/workflows/corpus-integrity-and-correction.
| Model | Runs | Steps finished | Found | Of those, right | Said so when it was missing |
|---|---|---|---|---|---|
| Claude Opus 5 | 5 | 96.0% | 75.8% | 48.0% | 100.0% |
| Claude Sonnet 5 | 5 | 100.0% | 64.2% | 31.4% | 100.0% |
| Claude Haiku 4.5 | not run | not run | not run | not run | not run |
| Claude Fable 5.1 | 5 | 100.0% | 71.7% | 48.7% | 100.0% |
Claude Haiku 4.5, not run: Before we buy a single run we check that a session can only read the files we put in front of it. That check never came back the way it has to, on any of the 3 tries we had written down beforehand, so we bought no runs at all. This is us declining to measure, and it says nothing about the model.
Run 5 times on each of 3 models. Of the 51 faults planted in a 25 page website, the models found between 73.7% and 100.0%. Pinned at commit a67dd34, run in Claude Code. Full results: openaddict.com/workflows/link-graph-and-metadata-parity-audit.
| Model | Runs | Steps finished | Found | Of those, right | Said so when it was missing |
|---|---|---|---|---|---|
| Claude Opus 5 | 5 | 100.0% | 92.9% | 82.0% | 100.0% |
| Claude Sonnet 5 | 5 | 100.0% | 73.7% | 84.3% | 100.0% |
| Claude Haiku 4.5 | not run | not run | not run | not run | not run |
| Claude Fable 5.1 | 5 | 100.0% | 100.0% | 98.9% | 100.0% |
Claude Haiku 4.5, not run: Before we buy a single run we check that a session can only read the files we put in front of it. That check never came back the way it has to, on any of the 3 tries we had written down beforehand, so we bought no runs at all. This is us declining to measure, and it says nothing about the model.
Run 5 times on each of 3 models. One thing was made to go wrong in 90 days of traffic, and the models named it correctly between 80.0% and 100.0% of the time. Pinned at commit a67dd34, run in Claude Code. Full results: openaddict.com/workflows/traffic-drop-triage.
Claude Haiku 4.5, not run: Before we buy a single run we check that a session can only read the files we put in front of it. That check never came back the way it has to, on any of the 3 tries we had written down beforehand, so we bought no runs at all. This is us declining to measure, and it says nothing about the model.
These were written by the people who run this site, and tested by them, against material they built. Nobody outside checked any of it.
- AI-Surface Visibility Baseline (FDM): fails connector readiness honestly; ships when citation-measurement has real runs.
- Attribution and Outcome Reporting (FDA): fails honest usability at template status; ships when the agreement log has real cross-property rows and post_merge_outcome data.
- Forward-deployed product manager track (FDPM): named to hold the track; not launched, and carried with the same visible not-launched treatment so it cannot read as an empty shipped track.
- CONNECTORS.md: the capability classes every workflow declares, the deliberate absences, and the operated-substrate boundary.
- AGREEMENT-LOG.md: the public schema for the agreement log the operated deployment writes, generalized from the canonical agreement-rate table.