The canonical phrase registry, tracked in git and invoked by key #408

New issue

Closed

opened 2026-08-13 12:10:30 +00:00 by coilyco-ops · 0 comments

coilyco-ops commented

2026-08-13 12:10:30 +00:00

Member

Slice of #176, which Kai approved and Delphi recorded at 07:06Z as "build it. This is the real fix."

Scope

agent/phrases.yaml holds canonical phrases. The model writes {{phrase:no-tool}} and the harness renders the tracked text in the blockquote-code form it already uses.

What the registry enforces, and why each rule exists

A phrase must survive the notice alphabet unchanged, checked at load rather than at reply time. A phrase that says one thing in git and another in the channel is worse than no registry.

Keys are lowercase and hyphenated, so they never need quoting and never look like prose.

Duplicates are refused. A shadowed phrase is one nobody can find by reading the file.

An unknown key is an error, not a rendered marker. {{phrase:typo}} reaching a member is worse than a failed turn: the failure is recoverable through the repair loop and the leaked marker is not recoverable once it is on screen.

Out of scope

Making the model use a key. Whether the keys exist and render correctly is separable from whether the model reaches for them, and the second is where the prompt still has a job. Shipping the registry first means the prompt half can be measured against something real.

Acceptance

The tracked registry loads, and every phrase in it renders inside the notice shape unaltered.
A key renders; an unknown key fails and names itself.
A reply with no invocation is byte-identical.
policy-check verifies the registry, so it cannot become another declared-but-unread artifact.

**Slice of** https://forgejo.coilysiren.me/coilyco-gaming/sirens-echo/issues/176, which Kai approved and Delphi recorded at 07:06Z as *"build it. This is the real fix."* ## Scope `agent/phrases.yaml` holds canonical phrases. The model writes `{{phrase:no-tool}}` and the harness renders the tracked text in the blockquote-code form it already uses. ## What the registry enforces, and why each rule exists **A phrase must survive the notice alphabet unchanged**, checked at load rather than at reply time. A phrase that says one thing in git and another in the channel is worse than no registry. **Keys are lowercase and hyphenated**, so they never need quoting and never look like prose. **Duplicates are refused.** A shadowed phrase is one nobody can find by reading the file. **An unknown key is an error, not a rendered marker.** `{{phrase:typo}}` reaching a member is worse than a failed turn: the failure is recoverable through the repair loop and the leaked marker is not recoverable once it is on screen. ## Out of scope **Making the model use a key.** Whether the keys exist and render correctly is separable from whether the model reaches for them, and the second is where the prompt still has a job. Shipping the registry first means the prompt half can be measured against something real. ## Acceptance - The tracked registry loads, and every phrase in it renders inside the notice shape unaltered. - A key renders; an unknown key fails and names itself. - A reply with no invocation is byte-identical. - policy-check verifies the registry, so it cannot become another declared-but-unread artifact.

coilyco-ops referenced this issue from a commit

2026-08-13 12:10:41 +00:00

feat(phrases): canonical phrases the model invokes by key

coilyco-ops referenced this issue from a pull request that will close it,

2026-08-13 12:11:09 +00:00

feat(phrases): canonical phrases the model invokes by key #409

coilyco-ops referenced this issue

2026-08-13 12:11:13 +00:00

Give the model a first-class way to invoke canonical phrases, tracked in git and rendered in the blockquote-code form #176

coilyco-ops closed this issue

2026-08-13 12:14:16 +00:00