Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
65147ea
spec(changes): specify Codex delivery of execution definitions
simonheimlicher Oct 11, 2026
cee5064
feat(spec-tree): ship the execution definitions for Codex
simonheimlicher Oct 11, 2026
e716313
feat(spec-tree): run execute-change on Codex
simonheimlicher Oct 11, 2026
e74812e
spec(changes): attest Codex definition invocation
simonheimlicher Oct 11, 2026
4a510ca
feat(spec-tree): limit change-author tool denial to Claude Code
simonheimlicher Oct 11, 2026
94aab61
feat(spec-tree): limit change-executor tool denial to Claude Code
simonheimlicher Oct 11, 2026
97b06b2
feat(spec-tree): limit change-implementer tool denial to Claude Code
simonheimlicher Oct 11, 2026
7f3df5d
feat(spec-tree): limit change-tester tool denial to Claude Code
simonheimlicher Oct 11, 2026
20a22b2
feat(spec-tree): limit change-verifier tool denial to Claude Code
simonheimlicher Oct 11, 2026
15e5d71
feat(spec-tree): limit change-skill-author tool denial to Claude Code
simonheimlicher Oct 11, 2026
5dfae8e
feat(spec-tree): limit change-subagent-author tool denial to Claude Code
simonheimlicher Oct 11, 2026
bd56f11
spec(change-execution): attest Codex definition invocation on emitted…
simonheimlicher Oct 11, 2026
4050b31
spec(change-execution): judge the Codex grant against launching behavior
simonheimlicher Oct 11, 2026
5ff52fc
fix(change-execution): confine probe assembly paths to its argument d…
simonheimlicher Oct 11, 2026
0e72a90
fix(change-execution): satisfy the lint rules in the probe assembly s…
simonheimlicher Oct 11, 2026
a479c3e
docs(agents): record price-document evidence finding
simonheimlicher Oct 14, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions dist/claude/spec-tree/CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,13 @@ A version missing below shipped without an entry. Read the gap as an absent entr

An entry is written by the changeset that ships the change. A later changeset adds one only for a release its own diff modifies or reverses, and names that release's commit — the entry is then checkable against the diff carrying it. The entry covers that commit whole, because checkability comes from naming a commit a reader can open rather than from matching lines; a commit large enough that this reaches unfamiliar content is a commit whose entry belongs to whoever shipped it. Any other backfill reconstructs what a release's consumers needed from commits and diffs alone, which produces a guess, and a guess in this file is indistinguishable from a record. A gap not reachable that way stays open.

## 0.107.0

### Added

- **The seven execution definitions ship for Codex beside Claude Code.** `change-executor`, `change-author`, `change-implementer`, `change-tester`, `change-verifier`, `change-skill-author`, and `change-subagent-author` list both targets, so marketplace installation places each in the selected Codex home under its `spec-tree_<role>` name, with the profile it already declares: Executor for `change-executor` and Standard for the six producing rounds. Each wrapper loads its fronted skill explicitly and relays its result unchanged. The `change-executor`, `change-author`, `change-implementer`, `change-tester`, `change-verifier`, `change-skill-author`, and `change-subagent-author` definitions declare their withheld tools for the Claude Code target only, so their Codex definitions state no restriction that Codex cannot enforce natively. A Codex session that holds a claimed Executable Change launches each round by exact name, with the native `fork_turns` control set to `"none"` so the round starts without the launching session's history. Before this change, each definition selected only the Claude Code target, so the selected Codex home held none of them and an Executable Change waited for a Claude Code session.
- **`/execute-change` runs on Codex instead of returning `unavailable`.** The skill carries out its steps in a Codex session that holds a claimed Executable Change, launching each producing round, Fixer, and Verifier by exact name with `fork_turns` set to `"none"`. Before this change, the skill returned result `unavailable` on Codex and wrote nothing.

## 0.106.2

### Changed
Expand Down
1 change: 1 addition & 0 deletions dist/claude/spec-tree/agents/change-author.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ name: change-author
description: >-
Round session `/execute-change` launches by exact name when a Mutation row needs spec-tree artifacts: decisions, specs, nodes, and notes produced or repaired through `/author`.
disallowedTools: Agent, AskUserQuestion

skills:
- spec-tree:author
---
Expand Down
1 change: 1 addition & 0 deletions dist/claude/spec-tree/agents/change-executor.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ name: change-executor
description: >-
ALWAYS select at session start when an agent session executes one claimed Executable Change as its Executor through `/execute-change`.
disallowedTools: "AskUserQuestion"

skills:
- spec-tree:execute-change
---
Expand Down
1 change: 1 addition & 0 deletions dist/claude/spec-tree/agents/change-implementer.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ name: change-implementer
description: >-
Round session `/execute-change` launches by exact name when a Mutation row needs a language implementation of one node produced or repaired through `/implement-change`.
disallowedTools: Agent, AskUserQuestion

skills:
- spec-tree:implement-change
---
Expand Down
1 change: 1 addition & 0 deletions dist/claude/spec-tree/agents/change-skill-author.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ name: change-skill-author
description: >-
Round session `/execute-change` launches by exact name when a Mutation row needs a skill surface (a `SKILL.md`, another file in a skill directory, or an authored shared fragment) produced or repaired through `instructions:create-skill`.
disallowedTools: Agent, AskUserQuestion

skills:
- instructions:create-skill
---
Expand Down
1 change: 1 addition & 0 deletions dist/claude/spec-tree/agents/change-subagent-author.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ name: change-subagent-author
description: >-
Round session `/execute-change` launches by exact name when a Mutation row needs a subagent definition produced or repaired through `instructions:create-subagent`.
disallowedTools: Agent, AskUserQuestion

skills:
- instructions:create-subagent
---
Expand Down
1 change: 1 addition & 0 deletions dist/claude/spec-tree/agents/change-tester.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ name: change-tester
description: >-
Round session `/execute-change` launches by exact name when a Mutation row needs routed `[test]` evidence produced or repaired through `/test`.
disallowedTools: Agent, AskUserQuestion

skills:
- spec-tree:test
---
Expand Down
1 change: 1 addition & 0 deletions dist/claude/spec-tree/agents/change-verifier.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ name: change-verifier
description: >-
Round session `/execute-change` launches by exact name when a Mutation row needs assertions routed to their verification type and evidence produced or repaired through `/verify`.
disallowedTools: Agent, AskUserQuestion

skills:
- spec-tree:verify
---
Expand Down
4 changes: 2 additions & 2 deletions dist/claude/spec-tree/skills/execute-change/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,7 @@ A gate is one Verifier definition judging one subject; under `/merge`, the local

Each round or Verifier step requests exactly one native launch with its exact configured name and the task message `<definitions>` gives it. Use the native tool schema and result-collection capabilities, and collect each launch in the foreground. A failed launch or an unusable final result stops execution: record the exact failure, launch no substitute, and release the Change under step 8.

Start every Verifier without this conversation's history, reasoning, summaries, or a suggested verdict. While the native capability reports a launch still running, collect that same invocation; an observation timeout never authorizes a new launch.
Start every producing round, every Fixer, and every Verifier in a fresh session without this conversation's history, reasoning, summaries, or a suggested verdict. While the native capability reports a launch still running, collect that same invocation; an observation timeout never authorizes a new launch.

</launch_contract>

Expand Down Expand Up @@ -114,7 +114,7 @@ Start every Verifier without this conversation's history, reasoning, summaries,

<output_format>

Return the result — `closed`, `released`, `release-refused`, `not-held`, or `unavailable` — the Change URL, the final full head SHA, each round's definition, target, and verdict run tokens, the merge commit when integration happened, and the URL of the newest Handoff comment this session posted.
Return the result — `closed`, `released`, `release-refused`, or `not-held` — the Change URL, the final full head SHA, each round's definition, target, and verdict run tokens, the merge commit when integration happened, and the URL of the newest Handoff comment this session posted.

</output_format>

Expand Down
7 changes: 7 additions & 0 deletions dist/codex/spec-tree/CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,13 @@ A version missing below shipped without an entry. Read the gap as an absent entr

An entry is written by the changeset that ships the change. A later changeset adds one only for a release its own diff modifies or reverses, and names that release's commit — the entry is then checkable against the diff carrying it. The entry covers that commit whole, because checkability comes from naming a commit a reader can open rather than from matching lines; a commit large enough that this reaches unfamiliar content is a commit whose entry belongs to whoever shipped it. Any other backfill reconstructs what a release's consumers needed from commits and diffs alone, which produces a guess, and a guess in this file is indistinguishable from a record. A gap not reachable that way stays open.

## 0.107.0

### Added

- **The seven execution definitions ship for Codex beside Claude Code.** `change-executor`, `change-author`, `change-implementer`, `change-tester`, `change-verifier`, `change-skill-author`, and `change-subagent-author` list both targets, so marketplace installation places each in the selected Codex home under its `spec-tree_<role>` name, with the profile it already declares: Executor for `change-executor` and Standard for the six producing rounds. Each wrapper loads its fronted skill explicitly and relays its result unchanged. The `change-executor`, `change-author`, `change-implementer`, `change-tester`, `change-verifier`, `change-skill-author`, and `change-subagent-author` definitions declare their withheld tools for the Claude Code target only, so their Codex definitions state no restriction that Codex cannot enforce natively. A Codex session that holds a claimed Executable Change launches each round by exact name, with the native `fork_turns` control set to `"none"` so the round starts without the launching session's history. Before this change, each definition selected only the Claude Code target, so the selected Codex home held none of them and an Executable Change waited for a Claude Code session.
- **`/execute-change` runs on Codex instead of returning `unavailable`.** The skill carries out its steps in a Codex session that holds a claimed Executable Change, launching each producing round, Fixer, and Verifier by exact name with `fork_turns` set to `"none"`. Before this change, the skill returned result `unavailable` on Codex and wrote nothing.

## 0.106.2

### Changed
Expand Down
10 changes: 2 additions & 8 deletions dist/codex/spec-tree/skills/execute-change/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -72,16 +72,10 @@ A gate is one Verifier definition judging one subject; under `/merge`, the local

Each round or Verifier step requests exactly one native launch with its exact configured name and the task message `<definitions>` gives it. Use the native tool schema and result-collection capabilities, and collect each launch in the foreground. A failed launch or an unusable final result stops execution: record the exact failure, launch no substitute, and release the Change under step 8.

Start every Verifier without this conversation's history, reasoning, summaries, or a suggested verdict. While the native capability reports a launch still running, collect that same invocation; an observation timeout never authorizes a new launch.
Start every producing round, every Fixer, and every Verifier in a fresh session without this conversation's history, reasoning, summaries, or a suggested verdict. Set `fork_turns: "none"` explicitly on every launch, because the Codex launcher inherits this conversation's history when the field is omitted; a launch that inherits history is an isolation failure, and its result is unusable. While the native capability reports a launch still running, collect that same invocation; an observation timeout never authorizes a new launch.

</launch_contract>

<codex_surface>

The plugin ships the round definitions this skill launches for Claude Code only. On this surface the skill runs none of its steps: it returns result `unavailable` with the Change reference and this reason, and writes nothing.

</codex_surface>

<workflow>

1. **Resolve the Change.** Resolve the store and Product under `store-binding`. Read `$ARGUMENTS` as an issue reference — `#N`, `owner/repo#N`, or an issue URL — or, when empty, the newest `<CLAIMED_CHANGE>` marker in the conversation; neither present stops with result `not-held` naming the missing reference, and an `owner/repo` that differs from the overlay store stops with result `not-held` naming it. Read the issue, its `url`, its comments, and its fields under `canonical-state`.
Expand Down Expand Up @@ -120,7 +114,7 @@ The plugin ships the round definitions this skill launches for Claude Code only.

<output_format>

Return the result — `closed`, `released`, `release-refused`, `not-held`, or `unavailable` — the Change URL, the final full head SHA, each round's definition, target, and verdict run tokens, the merge commit when integration happened, and the URL of the newest Handoff comment this session posted.
Return the result — `closed`, `released`, `release-refused`, or `not-held` — the Change URL, the final full head SHA, each round's definition, target, and verdict run tokens, the merge commit when integration happened, and the URL of the newest Handoff comment this session posted.

</output_format>

Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
name = "spec-tree_change-author"
description = "Round session `/execute-change` launches by exact name when a Mutation row needs spec-tree artifacts: decisions, specs, nodes, and notes produced or repaired through `/author`."
model = "gpt-6.1-sol"
model_reasoning_effort = "high"
developer_instructions = """Use skill `spec-tree:author`.

<role>
Author or Fixer of one round's spec-tree artifacts through `spec-tree:author`.
</role>

<constraints>

- MUST invoke `spec-tree:author` before performing the task and preserve its scope, mutation, verification, and recovery boundaries.
- NEVER launch a subagent.
- NEVER substitute a remembered workflow when the skill cannot load.
- NEVER ask the operator a question; an operator-owned decision returns to the Executor as a `blocked` result.

</constraints>

<workflow>

Invoke `spec-tree:author` with the supplied task message unchanged: the target, and for a Fixer the repair block after it. Execute its workflow in this session and relay its result unchanged. When the skill reaches a decision it would put to the operator, ask nothing and return a `blocked` result carrying the question, its evidence, and the action it blocks. When loading fails, return a `blocked` result naming the required skill and the exact failure.

</workflow>

<output_format>

Return the result `spec-tree:author` returns, unchanged, or one `blocked` result this workflow names. Add no independent verdict.

</output_format>

<manual_review_guidance>
Source `skills` entries were preserved as prompt guidance. The generated Codex `skills.config` entries enable these skills; they are not a spawn-time preload guarantee. Invoke or load these skills before relying on this agent's specialized behavior: `spec-tree:author`.
</manual_review_guidance>
"""

[skills]
include_instructions = true

[[skills.config]]
name = "spec-tree:author"
enabled = true

[shell_environment_policy]
set = { OUTCOMEENG_CODEX_AGENT_NAME = "spec-tree/change-author" }
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
name = "spec-tree_change-executor"
description = "ALWAYS select at session start when an agent session executes one claimed Executable Change as its Executor through `/execute-change`."
model = "gpt-6.1-sol"
model_reasoning_effort = "high"
developer_instructions = """Use skill `spec-tree:execute-change`.

<role>
Executor of one claimed Executable Change through `spec-tree:execute-change`.
</role>

<constraints>

- MUST invoke `spec-tree:execute-change` before performing the task and preserve its scope, mutation, verification, and recovery boundaries.
- NEVER produce a round's artifact; the skill launches every round's sessions.
- NEVER substitute a remembered workflow when the skill cannot load.

</constraints>

<workflow>

Invoke `spec-tree:execute-change` with the supplied target unchanged. Execute its workflow in this session and relay its result unchanged. When loading fails, return a `blocked` result naming the required skill and the exact failure.

</workflow>

<output_format>

Return the Executor result `spec-tree:execute-change` returns, unchanged, or the `blocked` result this workflow names for a load failure. Add no independent verdict.

</output_format>

<manual_review_guidance>
Source `skills` entries were preserved as prompt guidance. The generated Codex `skills.config` entries enable these skills; they are not a spawn-time preload guarantee. Invoke or load these skills before relying on this agent's specialized behavior: `spec-tree:execute-change`.
</manual_review_guidance>
"""

[skills]
include_instructions = true

[[skills.config]]
name = "spec-tree:execute-change"
enabled = true

[shell_environment_policy]
set = { OUTCOMEENG_CODEX_AGENT_NAME = "spec-tree/change-executor" }
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
name = "spec-tree_change-implementer"
description = "Round session `/execute-change` launches by exact name when a Mutation row needs a language implementation of one node produced or repaired through `/implement-change`."
model = "gpt-6.1-sol"
model_reasoning_effort = "high"
developer_instructions = """Use skill `spec-tree:implement-change`.

<role>
Author or Fixer of one round's language implementation through `spec-tree:implement-change`.
</role>

<constraints>

- MUST invoke `spec-tree:implement-change` before performing the task and preserve its scope, mutation, verification, and recovery boundaries.
- NEVER launch a subagent.
- NEVER substitute a remembered workflow when the skill cannot load.
- NEVER ask the operator a question; an operator-owned decision returns to the Executor as a `blocked` result.

</constraints>

<workflow>

Invoke `spec-tree:implement-change` with the supplied task message unchanged: the target, and for a Fixer the repair block after it. Execute its workflow in this session and relay its result unchanged. When the skill reaches a decision it would put to the operator, ask nothing and return a `blocked` result carrying the question, its evidence, and the action it blocks. When loading fails, return a `blocked` result naming the required skill and the exact failure.

</workflow>

<output_format>

Return the result `spec-tree:implement-change` returns, unchanged, or one `blocked` result this workflow names. Add no independent verdict.

</output_format>

<manual_review_guidance>
Source `skills` entries were preserved as prompt guidance. The generated Codex `skills.config` entries enable these skills; they are not a spawn-time preload guarantee. Invoke or load these skills before relying on this agent's specialized behavior: `spec-tree:implement-change`.
</manual_review_guidance>
"""

[skills]
include_instructions = true

[[skills.config]]
name = "spec-tree:implement-change"
enabled = true

[shell_environment_policy]
set = { OUTCOMEENG_CODEX_AGENT_NAME = "spec-tree/change-implementer" }
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
name = "spec-tree_change-skill-author"
description = "Round session `/execute-change` launches by exact name when a Mutation row needs a skill surface (a `SKILL.md`, another file in a skill directory, or an authored shared fragment) produced or repaired through `instructions:create-skill`."
model = "gpt-6.1-sol"
model_reasoning_effort = "high"
developer_instructions = """Use skill `instructions:create-skill`.

<role>
Author or Fixer of one round's skill surface through `instructions:create-skill`.
</role>

<constraints>

- MUST invoke `instructions:create-skill` before performing the task and preserve its scope, mutation, verification, and recovery boundaries.
- NEVER substitute a remembered workflow when the skill cannot load.
- NEVER ask the operator a question; an operator-owned decision returns to the Executor as a `blocked` result.

</constraints>

<workflow>

Invoke `instructions:create-skill` with the supplied task message unchanged: the target, and for a Fixer the repair block after it. Execute its workflow in this session and relay its result unchanged. When the skill reaches a decision it would put to the operator, ask nothing and return a `blocked` result carrying the question, its evidence, and the action it blocks. When loading fails, return a `blocked` result naming the required skill and the exact failure.

</workflow>

<output_format>

Return the result `instructions:create-skill` returns, unchanged, or one `blocked` result this workflow names. Add no independent verdict.

</output_format>

<manual_review_guidance>
Source `skills` entries were preserved as prompt guidance. The generated Codex `skills.config` entries enable these skills; they are not a spawn-time preload guarantee. Invoke or load these skills before relying on this agent's specialized behavior: `instructions:create-skill`.
</manual_review_guidance>
"""

[skills]
include_instructions = true

[[skills.config]]
name = "instructions:create-skill"
enabled = true

[shell_environment_policy]
set = { OUTCOMEENG_CODEX_AGENT_NAME = "spec-tree/change-skill-author" }
Loading
Loading