# Adapter correspondence

## Confirmed ecosystem correspondence

- **Status:** confirmed correspondence
- **Paper wording:** Strong correspondence. Parity requires runtime tests.
- **Case:** gstack Skill suite (`garrytan/gstack`)
- **Fixed upstream commit:** `11de390be1be6849eb9a15f91ff4922dd16c589a`
- **Canonical Adaptee template:**
  [`SKILL.md.tmpl`](https://github.com/garrytan/gstack/blob/11de390be1be6849eb9a15f91ff4922dd16c589a/SKILL.md.tmpl)
- **Generated Skill surface:**
  [`SKILL.md`](https://github.com/garrytan/gstack/blob/11de390be1be6849eb9a15f91ff4922dd16c589a/SKILL.md)
- **Generation implementation:**
  [`scripts/gen-skill-docs.ts`](https://github.com/garrytan/gstack/blob/11de390be1be6849eb9a15f91ff4922dd16c589a/scripts/gen-skill-docs.ts)
- **Claude target configuration:**
  [`hosts/claude.ts`](https://github.com/garrytan/gstack/blob/11de390be1be6849eb9a15f91ff4922dd16c589a/hosts/claude.ts)
- **Codex target configuration:**
  [`hosts/codex.ts`](https://github.com/garrytan/gstack/blob/11de390be1be6849eb9a15f91ff4922dd16c589a/hosts/codex.ts)
- **Codex invocation evidence:**
  [`test/codex-e2e.test.ts`](https://github.com/garrytan/gstack/blob/11de390be1be6849eb9a15f91ff4922dd16c589a/test/codex-e2e.test.ts)
- **Vendored observation record:**
  [frozen evidence](evidence/gstack-frozen-case.md)

`SKILL.md.tmpl` is the canonical Adaptee template. The generated `SKILL.md`
surface is output, not the canonical source. The generator plus a selected
system-specific binding is the Adapter because it translates discovery paths,
frontmatter, tool names, and other target conventions without making the
canonical procedure target-specific. The Host-specific Skill contract is the
Target, and task-level invocation is the Client. An Agent Host can carry and
activate the binding, while the Agent Runtime interprets it; neither platform
role becomes a GoF participant automatically.

The local evidence record preserves the fixed revision, exact paths,
counterevidence, and claim boundary needed to audit this correspondence without
access to a separate research repository. The controlled status is confirmed
correspondence; the separately labeled paper wording records the qualification
that runtime parity still requires tests under each Target Host.

## Constructive sample

- **Status:** constructive
- **Unit:** [`sample/SKILL.md`](sample/SKILL.md) with three tracker bindings
- **Target contract:**
  [`sample/references/tracker-contracts.md`](sample/references/tracker-contracts.md)
- **Executable oracle:** [`sample/scripts/run_demo.py`](sample/scripts/run_demo.py)
- **Focused verification:** [`sample/tests/test_demo.py`](sample/tests/test_demo.py)

The local multi-tracker sample demonstrates that semantic translation can be
built and tested. It is not evidence about gstack, open-source frequency, or
comparative quality. Conversely, the gstack case does not validate the local
sample; each claim has its own artifacts and verification boundary.