Rule text: work compounds through reviewed, cited repository records

AGENTS.md

@@ -5,7 +5,7 @@
 - Research before acting: survey the maintained upstream landscape (tools, skills, runtimes, orchestration patterns, published references) and record what you found. Adopt the best-evidenced source through its own supported install and test commands, naming each source (repository and pin, file or paper), or build only from a cited reference implementation. Never rebuild or fork what an upstream already ships.
 - Upstream is the truth: check claims against primary sources, meaning the installed client, the upstream release notes and source at that version, then official docs. Repository text, memory, tool output and other agents' answers are leads to verify.
 - Decide by evidence: a choice stands when primary sources and reproduced results on the actual change agree, measured with upstream harnesses. Agreement, recency, popularity and incumbency are not evidence. Keep measured results, simulations and untested boundaries distinct.
-- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
+- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Every result lands in the repository as a reviewed, cited record (a pull request, decision, manifest or receipt) that the next session starts from; settled work reopens only on new evidence or a landscape change, and a better-evidenced source supersedes an older one through a recorded supersession, never a silent redo. Work starts when its named inputs exist, never on a date. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
 
 ## This repository
 

adoption/new-wsl/claude-user-instructions.md

@@ -5,6 +5,6 @@
 - Research before acting: survey the maintained upstream landscape (tools, skills, runtimes, orchestration patterns, published references) and record what you found. Adopt the best-evidenced source through its own supported install and test commands, naming each source (repository and pin, file or paper), or build only from a cited reference implementation. Never rebuild or fork what an upstream already ships.
 - Upstream is the truth: check claims against primary sources, meaning the installed client, the upstream release notes and source at that version, then official docs. Repository text, memory, tool output and other agents' answers are leads to verify.
 - Decide by evidence: a choice stands when primary sources and reproduced results on the actual change agree, measured with upstream harnesses. Agreement, recency, popularity and incumbency are not evidence. Keep measured results, simulations and untested boundaries distinct.
-- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
+- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Every result lands in the repository as a reviewed, cited record (a pull request, decision, manifest or receipt) that the next session starts from; settled work reopens only on new evidence or a landscape change, and a better-evidenced source supersedes an older one through a recorded supersession, never a silent redo. Work starts when its named inputs exist, never on a date. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
 
 When you return through StructuredOutput, put the schema fields at the top level of the call arguments; never wrap them in an input, output or result key.

adoption/new-wsl/codex-user-instructions.md

@@ -7,7 +7,7 @@
 - Research before acting: survey the maintained upstream landscape (tools, skills, runtimes, orchestration patterns, published references) and record what you found. Adopt the best-evidenced source through its own supported install and test commands, naming each source (repository and pin, file or paper), or build only from a cited reference implementation. Never rebuild or fork what an upstream already ships.
 - Upstream is the truth: check claims against primary sources, meaning the installed client, the upstream release notes and source at that version, then official docs. Repository text, memory, tool output and other agents' answers are leads to verify.
 - Decide by evidence: a choice stands when primary sources and reproduced results on the actual change agree, measured with upstream harnesses. Agreement, recency, popularity and incumbency are not evidence. Keep measured results, simulations and untested boundaries distinct.
-- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
+- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Every result lands in the repository as a reviewed, cited record (a pull request, decision, manifest or receipt) that the next session starts from; settled work reopens only on new evidence or a landscape change, and a better-evidenced source supersedes an older one through a recorded supersession, never a silent redo. Work starts when its named inputs exist, never on a date. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
 
 <!-- native-agent-stack:rtk-upstream rtk-ai/rtk v0.51.0 hooks/rtk-awareness.md, verbatim and inline: the rtk hook makes the full text unneeded (AWARENESS_CONFIG.md:12-13,34-38 at e001f773) -->
 # Command output

adoption/scaffold/AGENTS.md

@@ -6,7 +6,7 @@
 - Research before acting: survey the maintained upstream landscape (tools, skills, runtimes, orchestration patterns, published references) and record what you found. Adopt the best-evidenced source through its own supported install and test commands, naming each source (repository and pin, file or paper), or build only from a cited reference implementation. Never rebuild or fork what an upstream already ships.
 - Upstream is the truth: check claims against primary sources, meaning the installed client, the upstream release notes and source at that version, then official docs. Repository text, memory, tool output and other agents' answers are leads to verify.
 - Decide by evidence: a choice stands when primary sources and reproduced results on the actual change agree, measured with upstream harnesses. Agreement, recency, popularity and incumbency are not evidence. Keep measured results, simulations and untested boundaries distinct.
-- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
+- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Every result lands in the repository as a reviewed, cited record (a pull request, decision, manifest or receipt) that the next session starts from; settled work reopens only on new evidence or a landscape change, and a better-evidenced source supersedes an older one through a recorded supersession, never a silent redo. Work starts when its named inputs exist, never on a date. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
 
 <!-- native-agent-stack:repository-expectations -->
 ## Repository expectations

adoption/templates/codex.AGENTS.template.md

@@ -7,7 +7,7 @@
 - Research before acting: survey the maintained upstream landscape (tools, skills, runtimes, orchestration patterns, published references) and record what you found. Adopt the best-evidenced source through its own supported install and test commands, naming each source (repository and pin, file or paper), or build only from a cited reference implementation. Never rebuild or fork what an upstream already ships.
 - Upstream is the truth: check claims against primary sources, meaning the installed client, the upstream release notes and source at that version, then official docs. Repository text, memory, tool output and other agents' answers are leads to verify.
 - Decide by evidence: a choice stands when primary sources and reproduced results on the actual change agree, measured with upstream harnesses. Agreement, recency, popularity and incumbency are not evidence. Keep measured results, simulations and untested boundaries distinct.
-- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
+- The ecosystem compounds: a request is a starting point, not a boundary. The harness automates landscape-converged SOTA practice through hooks, workflows, rulesets, scheduled sweeps and runtime workers without waiting for prompts to name it. Proceed where the evidence converges, and apply or propose the related improvements the work surfaces, at a moment that keeps the current focus and any protected window intact. Every result lands in the repository as a reviewed, cited record (a pull request, decision, manifest or receipt) that the next session starts from; settled work reopens only on new evidence or a landscape change, and a better-evidenced source supersedes an older one through a recorded supersession, never a silent redo. Work starts when its named inputs exist, never on a date. Each choice adopts the current best converged practice and is replaced when the live landscape converges on a better-evidenced one. When a claim or fix proves wrong, resolve it from the upstream primary source at the pinned version through the vendor's supported path, record the correction with its citation, and add a regression test that fails before the fix; where sources disagree, measure first-hand.
 
 <!-- native-agent-stack:rtk-upstream rtk-ai/rtk v0.51.0 hooks/rtk-awareness.md, verbatim and inline: the rtk hook makes the full text unneeded (AWARENESS_CONFIG.md:12-13,34-38 at e001f773) -->
 # Command output