feat(re-anchor): add script-the-deterministic-work corrector (0.3.0)

[kyle-sexton]

Summary

Adds a sixth discipline corrector to the re-anchor plugin:
/re-anchor:script-the-deterministic-work. It re-anchors the discipline
that purely deterministic sub-work â^@^T counting, diffing, sorting,
transforming, matching, sweeping across files, arithmetic â^@^T gets a
script that runs and returns real output, and the model reasons only
afterward over that output. Then it audits the work in flight for
transforms executed by hand and re-derives them from a script's result.

Firing it is a re-anchor, not an accusation; the audit may return clean.
The skill is a thin delta over the shared re-anchor/audit/correct engine
(plugins/re-anchor/context/re-anchor-audit-correct.md) â^@^T it adds only
what is specific to scripting deterministic work.

The tier boundary (cited, not restated)

The boundary of what to script re-anchors the consuming org's
enforceability-tiers.md convention, used as the boundary tool so the
skill does not degrade into "automate everything":

• Deterministic â^F^R script it, run it, reason over the output.
• Detect-then-judge â^F^R script only the detect half; the verdict
  stays model judgement (a flag is a candidate, never the ruling).
• Reasoning-only â^F^R never script it; a script here manufactures false
  confidence.

Flagged standards gap

enforceability-tiers.md classifies conventions by tier and routes a
recurring finding to the mechanism its tier permits. It does not
cover the in-task move this skill re-anchors â^@^T "this task needs a count
or a diff now, so script it now." That application has no dedicated
standards convention, so the skill flags the gap rather than inventing a
rubric, exactly as tighten-your-output flags its markdown-terseness gap
(standards#164). Companion issue filed: melodic-software/standards#169.

Distinct from standing automation

A recurring deterministic check routes to a standing hook (a
config/automation-gaps capability). This corrector owns the complementary
one-off, session-time case: make a script now, often throwaway,
feed it the input, reason over its output.

Changes

• New skill skills/script-the-deterministic-work/ (SKILL.md + 4 evals
  mirroring the sibling shape).
• plugin.json: version 0.1.0 â^F^R 0.2.0 (added skill = minor bump,
  matching the session-flow keep-going precedent), description, keywords.
• README table + section; CHANGELOG [0.2.0]; marketplace tag.
• Root catalog regenerated via scripts/generate-catalog.mjs (never
  hand-edited).

Verification

• markdownlint-cli2 (MD013=80): clean.
• skill-quality static check: PASS â^@^T 0 errors, 0 warnings.
• evals conform to the bundled schema.
• generate-catalog.mjs --check: in sync.
• validate-plugin-contracts.mjs: passed.
• claude plugin validate --strict: passed.

Related

No linked issue — the work was contracted via the in-repo
repeated-prompts-to-skills PLAN (Round 2, corrector #6), not a
pre-existing ticket. Companion standards gap filed as
melodic-software/standards#169. Name locked
script-the-deterministic-work via the blind-generator naming protocol.

ð^_¤^V Generated with Claude Code

Contracted as corrector #6 of the re-anchor Round 2 grouping.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c1c5f3e183

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude finished @kyle-sexton's task in 2m 10s —— View job

---

PR Review — feat(re-anchor): add script-the-deterministic-work corrector (0.2.0)

• Gather context
• Read changed files
• Analyze skill content and structure
• Check alignment with plugin conventions
• Post review findings

---

Summary

This is a clean, well-structured addition that follows the established re-anchor plugin shape faithfully. The skill is correctly scoped to the one-off, session-time case; the tier vocabulary is well-documented and honestly flags the standards gap; the evals cover the four key behavior dimensions. One trigger phrase needs narrowing before merge; the rest is minor.

---

Findings

P2 — Trigger phrase 'script it' is too broad

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md (frontmatter description field)

Codex flagged this and the concern holds. 'script it' is a common code-authoring phrase — a user saying "can you script it?" almost always means "write a script," not "fire the deterministic-work corrector." Every sibling trigger is discipline-specific: 'tighten your output', 'too verbose', 'don't assume', 'stop hand-tallying' — none collapse to a single generic verb. 'script it' does.

The risk: with the plugin installed, a user asking Claude to "script it" (write them a bash or Python file) gets the drift-corrector instead of the script.

Recommended fix: Remove 'script it' from the trigger list. The remaining phrases ('script the deterministic work', 'don't eyeball that', 'you counted that by hand', 'compute that, don't estimate', 'diff it with a tool', 'stop hand-tallying', 'run it instead of guessing') already cover all the legitimate re-anchor invocation patterns without ambiguity. Fix this →

---

P3 — Gotcha header reads awkwardly

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L114

"Reason after over results" here means where the computation happens

Two prepositions in a row ("after over") with different semantic roles (temporal / about) makes the header dense to parse on first read. The body is clear, but the header should introduce what the body clarifies. Consider: "'Reason after' means where the computation happens" or "'Reason over output' vs. 'reason over inherited content'" to distinguish it from the sibling skill immediately. Minor — not blocking.

---

What's Good

• Structural fidelity. Correctly delegates to re-anchor-audit-correct.md for the shared method and adds only what is discipline-specific — the same pattern as every sibling skill. No method leakage.
• Tier vocabulary. The three-tier shape (deterministic / detect-then-judge / reasoning-only) is clearly stated, the sourcing ladder is explicit, and the portable-baseline fallback is present.
• Standards gap handling. Flags the in-task scripting gap honestly and cites the tighten-your-output precedent for the pattern. Does not invent a rubric.
• Scope boundary. The "distinct from standing automation" section makes the scope crisp: recurring → hook, session-time one-off → this skill. The eval for this case (ci: onboard to the CI platform (ci-workflows + standards) #4) is concrete and tests the routing behavior.
• Evals. All four cover distinct behaviors: happy path, over-application guard, gap-flagged-not-invented, and the standing-automation routing. The expected outputs are testable.
• Metadata. Version bump (0.1.0 → 0.2.0) is correct for an additive skill. plugin.json description is accurate, keywords extended appropriately. Marketplace tag and catalog regeneration are in sync.
• Gotchas section. The detect-then-judge trap (a flag is a shortlist, not a ruling) is the subtle miss — calling it out is exactly right.

---

The P2 trigger fix is the only change I'd want before merge.

[claude]

Claude finished @kyle-sexton's task in 1m 36s —— View job

---

PR Review — feat(re-anchor): cover the generation direction in script-the-deterministic-work

• Gather context
• Read changed files from latest commit
• Check whether previous P2 finding was addressed
• Post updated review findings

---

What this commit adds

The generation direction (b0daefc) is a coherent, well-scoped extension: deterministic structure (PR bodies, issue bodies, config boilerplate, skill skeletons) belongs in a script or native template; model output belongs in the judgment slots. The "prefer a native mechanism" note (repo PR/issue templates emit the scaffold at no cost) is exactly the right practical nuance. The fifth eval tests the generation path correctly, and the plugin.json description is updated accurately.

---

P2 — 'script it' trigger still present (unaddressed)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md (frontmatter, line 3)

The concern from the initial review was not addressed in this push. 'script it' remains in the trigger list alongside the unambiguous phrases. Every sibling corrector's trigger is discipline-specific; this one collapses to a generic verb that matches ordinary "write a script for me" requests.

Recommended fix: Remove 'script it' from the description frontmatter. The remaining phrases ('script the deterministic work', 'don't eyeball that', 'you counted that by hand', 'compute that, don't estimate', 'diff it with a tool', 'stop hand-tallying', 'run it instead of guessing') already cover all the legitimate re-anchor invocation patterns without ambiguity. Fix this →

---

P3 — Gotcha header still reads awkwardly (pre-existing, minor)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L129

"Reason after over results" here means where the computation happens

Unchanged from the initial review — "after over" with different semantic roles (temporal / about) in sequence makes the header dense on first read. Not blocking.

---

What's Good

• Generation direction is well-scoped. The subsection correctly extends the same principle — reserve model output for judgment, emit structure from a template — without blurring the core "script and run" discipline. The "prefer a native mechanism" note keeps it practical and prevents over-engineering.
• Audit bullet. The scaffold bullet is concrete and lists real-world examples (PR body, issue, report, config boilerplate). It mirrors the precision of the existing analysis bullets.
• Eval feat(hook-telemetry): marketplace-wide telemetry contract + markdown-formatter producer #5. Tests the generation path cleanly: the prompt is a realistic invocation, the expected output names the right behavior (emit scaffold from template, fill judgment slots with model output), and the three expectations are independently testable.
• Plugin.json description. Correctly broadened to cover both directions ("offload deterministic sub-work — counts, diffs, sorts, transforms, and scaffolds — to a script or template, reserving model output for judgment").

---

The P2 trigger fix is the only change needed before merge. The generation extension itself is clean.

[claude]

Claude finished @kyle-sexton's task in 1m 17s —— View job

---

PR Review — feat(re-anchor): add script-the-deterministic-work corrector (0.2.0)

• Gather context
• Read changed files
• Check whether P2 finding ('script it' trigger) was addressed in this sync
• Check P3 gotcha header
• Post final review findings

---

Summary (third pass)

The PR is clean except for the one blocker that has now survived three review cycles unaddressed: 'script it' remains in the trigger list. Everything added in this sync — the generation direction, eval #5, the updated plugin description — is well-scoped and correct. One removal and this is ready to merge.

---

P2 — 'script it' trigger still present (third pass, still blocking)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L3

'script it' is still in the description frontmatter trigger list. The concern has been stated twice: it is a generic verb phrase — "can you script it?" is a routine code-authoring request — while every sibling trigger is discipline-specific:

Skill	Shortest trigger
do-your-research	"cite that"
follow-our-standards	"follow the standards"
point-dont-copy	"you copied that"
reason-dont-recite	"you're deferring to precedent"
tighten-your-output	"tighten this"
script-the-deterministic-work	"script it" ← generic

With the plugin installed, a user asking Claude to "script it" (meaning: write them a bash or Python script) fires the drift corrector instead. The remaining triggers ('script the deterministic work', 'don't eyeball that', 'you counted that by hand', 'compute that, don't estimate', 'diff it with a tool', 'stop hand-tallying', 'run it instead of guessing') already cover every legitimate re-anchor invocation without ambiguity.

Fix: Remove 'script it' from the description trigger list on line 3. Fix this →

---

P3 — Gotcha header density (pre-existing, non-blocking)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L129

"Reason after over results" here means where the computation happens

"After over" with two different semantic roles (temporal / about) in sequence makes the header dense on first read. The body is clear; the header should introduce what the body clarifies. Not blocking.

---

What's good

• Generation direction. The b0daefc extension is well-scoped: deterministic structure (PR bodies, issue bodies, config boilerplate, skill skeletons) belongs in a script or native template; model output belongs in the judgment slots. The "prefer a native mechanism" note is the right practical nuance. No method leakage.
• Eval feat(hook-telemetry): marketplace-wide telemetry contract + markdown-formatter producer #5. Tests the generation path cleanly: prompt is realistic, expected output names the right behavior (emit scaffold from template, fill judgment slots with model output), three expectations are independently testable.
• Tier vocabulary. Three-tier shape is clearly stated, sourcing ladder is explicit, portable-baseline fallback is present for consuming projects with no such convention.
• Standards gap handling. Flags the in-task scripting gap honestly; does not invent a rubric. Cites the tighten-your-output precedent pattern correctly.
• Scope boundary. "Distinct from standing automation" makes the one-off vs. recurring distinction crisp; eval ci: onboard to the CI platform (ci-workflows + standards) #4 tests the routing behavior concretely.
• Metadata. Version bump (0.1.0 → 0.2.0) is correct for an additive skill. CHANGELOG entry is accurate and well-structured. Keywords extended appropriately. Catalog regenerated.

---

Remove 'script it' and this is ready to merge.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 796a892e64

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude finished @kyle-sexton's task in 1m 56s —— View job

---

PR Review — feat(re-anchor): add script-the-deterministic-work corrector (0.2.0)

• Gather context
• Read changed files (latest HEAD)
• Check P2: 'script it' trigger resolved? No — still present (4th consecutive review)
• Check Codex P2: config/automation-gaps route
• Check P3: gotcha header
• Post final review findings

---

Summary (fourth pass)

The PR remains one line from merge. 'script it' has survived four consecutive review cycles without being removed. A second finding — a cross-plugin routing reference that doesn't name the concrete invocable skill — is confirmed against the codebase. Both are fixable in the same file in under a minute.

---

P2 — 'script it' trigger still present (fourth pass, still blocking)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L3

'script it' is still in the description frontmatter trigger list on line 3. This has been flagged in every review pass (initial Claude review, Codex review, second Claude pass, third Claude pass, now fourth). The phrase is a generic verb — "can you script it?" is a routine code-authoring request — while every sibling trigger is discipline-specific:

Skill	Shortest trigger
do-your-research	'cite that'
follow-our-standards	'follow the standards'
point-dont-copy	'you copied that'
reason-dont-recite	'you're deferring to precedent'
tighten-your-output	'tighten this'
script-the-deterministic-work	'script it' ← generic

With the plugin installed, a user asking Claude to "script it" (meaning: write them a bash or Python script) fires the drift corrector instead. The remaining triggers — 'script the deterministic work', 'don't eyeball that', 'you counted that by hand', 'compute that, don't estimate', 'diff it with a tool', 'stop hand-tallying', 'run it instead of guessing' — already cover every legitimate re-anchor invocation pattern without ambiguity.

Fix: Remove 'script it' from the description trigger list on line 3. Fix this →

---

P2 — Cross-plugin routing reference doesn't name the concrete invocable skill

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L105

That is the territory of a config/automation-gaps capability: institutionalize the check so it never reaches review again.

config/automation-gaps is not an invocable command. The concrete skill is /claude-config:automation-gaps, which exists in this repo at plugins/claude-config/skills/automation-gaps/. The codebase-health plugin handles this correctly — its SKILL.md says: "if the claude-config plugin is installed, route those to /claude-config:audit and /claude-config:automation-gaps; otherwise state they are out of scope." The re-anchor skill uses a vague label that the model cannot act on when the plugin is installed.

The same vague reference appears in eval #4's expected_output and expectations fields in evals/evals.json#L44.

Fix: In SKILL.md line 105, replace a config/automation-gaps capability with \/claude-config:automation-gaps` (when the claude-config plugin is installed). In evals.json, update the expected_outputandexpectationsstrings in eval #4 to match — e.g.,"Routes the recurring rule to `/claude-config:automation-gaps` (when installed) rather than absorbing it"`. Fix this →

---

P3 — Gotcha header density (pre-existing, non-blocking)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L129

"Reason after over results" here means where the computation happens

"After over" with two different semantic roles (temporal / about) in sequence makes the header dense on first read. The body is clear; the header should introduce what it clarifies. Not blocking.

---

What's good

• Structural fidelity. Correctly delegates to re-anchor-audit-correct.md and adds only discipline-specific content. No method leakage.
• Tier vocabulary. Three-tier shape (deterministic / detect-then-judge / reasoning-only) is well-stated; sourcing ladder and portable-baseline fallback are both present.
• Generation direction. The b0daefc extension — deterministic structure emits from a script or native template, model output reserves for judgment slots — is well-scoped. "Prefer a native mechanism" is the right practical nuance.
• Standards gap handling. Honest about the missing in-task-scripting convention; cites the tighten-your-output precedent for the gap-flagging pattern; does not invent a rubric.
• Scope boundary. "Distinct from standing automation" makes the one-off vs. recurring split crisp; eval ci: onboard to the CI platform (ci-workflows + standards) #4 tests the routing behavior concretely.
• Evals. Five evals covering distinct behaviors: happy path, over-application guard, gap-flagged-not-invented, standing-automation routing, generation path. All expectations are independently testable.
• Metadata. Version bump (0.1.0 → 0.2.0) is correct; CHANGELOG entry accurate; keywords extended; marketplace tag and catalog in sync.

---

Two fixes — remove 'script it' from line 3, update the config/automation-gaps reference to name the concrete /claude-config:automation-gaps skill — and this is ready to merge.

[claude]

Claude finished @kyle-sexton's task in 1m 28s —— View job

---

PR Review — feat(re-anchor): add script-the-deterministic-work corrector (0.2.0)

• Gather context
• Read latest SKILL.md and evals.json
• Verify P2 fixes (trigger phrase + capability name)
• Check P3 gotcha header
• Post final review findings

---

Summary (fifth pass)

Both blocking P2 findings from the previous four cycles are resolved in this commit. This is ready to merge.

---

P2a — 'script it' trigger: FIXED

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L3

'script it' is gone from the trigger list. It's been replaced by 'you should have scripted that' (discipline-specific and retrospective, not a work-order phrase) and a disambiguation note: "Not for authoring a requested script or migration ('script it' as a work order is script-writing, not this corrector)." That note is better than a simple removal — it actively disambiguates for a model reading the frontmatter, which prevents the hijack without leaving a gap a consumer might wonder about.

---

P2b — Capability name: FIXED in SKILL.md; minor residue in evals.json

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L105

SKILL.md now reads:

an automation-gaps capability (/claude-config:automation-gaps when that plugin is installed; prose guidance otherwise)

That's the correct pattern — concrete invocable name, conditional on installation, prose fallback. Matches how codebase-health handles the same cross-plugin reference.

The expectations string in eval #4 still reads "Routes the recurring rule to a config/automation-gaps capability rather than absorbing it" (evals.json#L48) — the vague form, not the concrete skill name. This is minor (an eval expectation tests conceptual behavior, not exact prose), but it's inconsistent with the SKILL.md fix. Not blocking. Fix this →

---

P3 — Gotcha header density (pre-existing, non-blocking)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L130

"Reason after over results" here means where the computation happens

Unchanged. "After over" with two different semantic roles (temporal / about) still reads dense. Not blocking.

---

What's Good

• Trigger fix is better than a removal. 'you should have scripted that' is discipline-specific and retrospective (exactly right for a corrector); the disambiguation note proactively handles the 'script it' ambiguity without leaving a gap.
• Capability reference pattern is correct. Named invocable + installation conditional + prose fallback — matches the codebase convention for optional cross-plugin routing.
• Structural fidelity, tier vocabulary, generation direction, evals, and metadata: all unchanged and correct. Everything noted as good in prior passes still holds.

---

Both blocking P2 findings are resolved. The evals.json note is the only residue and it is not blocking. Merge when ready.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a8e48d72ad

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude finished @kyle-sexton's task in 1m 52s —— View job

---

PR Review — feat(re-anchor): add script-the-deterministic-work corrector (0.2.0)

• Gather context
• Read changed files (latest HEAD: 9c52a2c)
• Verify Codex P2: first-turn auto-invoke fix
• Check evals.json residue (minor, from pass 5)
• Check P3 gotcha header
• Post final review findings

---

Summary (sixth pass)

All blocking findings from every prior cycle are resolved. This is ready to merge.

---

Codex P2 — First-turn auto-invoke: FIXED

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L3

The 9c52a2c commit adds the correct exclusion to the description frontmatter:

"…and not for a first-turn count/diff work order ('diff these files', 'count the routes') — doing that task with a tool is just doing the task; this corrector fires on drift, mid-flight or retrospective, or as posture at the start of transform-heavy work."

This is a better fix than disable-model-invocation: true would have been. Setting that flag would have killed the legitimate "posture at start of transform-heavy work" use case; the prose exclusion distinguishes a work order ("diff these files") from transform-heavy-session posture — precisely the distinction needed. The description now clearly scopes the corrector to (a) mid-flight/retrospective drift and (b) posture on a transform-heavy task already in motion.

---

Residual (non-blocking) — Evals.json eval #4 expectations string

File: plugins/re-anchor/skills/script-the-deterministic-work/evals/evals.json#L48

"Routes the recurring rule to a config/automation-gaps capability rather than absorbing it"

The SKILL.md body was updated (in a8e48d7) to name /claude-config:automation-gaps concretely; this eval expectation still uses the vague form. Noted previously as non-blocking — an eval expectation tests conceptual routing behavior, not exact prose — but it's inconsistent with the fix applied upstream. Worth a follow-up ticket or a fast-follow PR; not a blocker. Fix this →

---

P3 — Gotcha header density (pre-existing, non-blocking)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L130

"Reason after over results" here means where the computation happens

Unchanged across all six passes. "After over" with two different semantic roles in sequence reads dense on first sight. Not blocking.

---

What's Good

• All blocking findings cleared. Six review cycles, three from Claude and three from Codex; every P2 finding is resolved.
• First-turn fix is well-scoped. The exclusion uses concrete examples ('diff these files', 'count the routes') and articulates the right principle ("doing that task with a tool is just doing the task"), which a model can act on without ambiguity. Better than blunt disable-model-invocation.
• Trigger disambiguations now form a complete picture. The frontmatter description excludes both work-order script-writing ('script it') and first-turn work orders ('diff these files'), and the disambiguation note added in a8e48d7 remains. A model reading the description has everything needed to avoid false positives.
• Everything from prior passes still holds: structural fidelity, tier vocabulary, generation direction, scope boundary, five well-scoped evals, correct version bump and metadata.

---

Ready to merge.

[claude]

Claude finished @kyle-sexton's task in 3m 10s —— View job

---

PR Review — feat(re-anchor): add script-the-deterministic-work corrector (0.3.0)

• Gather context
• Read changed files at current HEAD
• Check what changed in the merge commit (main → branch, 0.3.0)
• Verify residual findings from prior passes
• Post final review

---

Summary (seventh pass)

All six prior blocking findings are resolved. This pass surfaces one new blocking defect introduced in the 0.3.0 merge commit: a missing closing code fence in plugins/re-anchor/README.md that swallows the entire script-the-deterministic-work section — heading and description — inside the mind-your-maxims code block. Also clearing the non-blocking evals.json residual from pass 5 and 6.

---

P1 — Missing closing code fence swallows the new section in README.md

File: plugins/re-anchor/README.md#L202

The ```shell block for mind-your-maxims (opened at line 202) is never closed before the ### script-the-deterministic-work heading is inserted at line 204. In GitHub Flavored Markdown, ```shell is not a valid closing fence — only a bare ``` (no trailing text) closes a fenced code block. The closing ``` at line 224 closes the first block, so the rendered output is one large code block containing the mind-your-maxims command, the ### script-the-deterministic-work heading, the full description paragraph, and the script-the-deterministic-work command — all rendered as literal text.

As rendered on GitHub:

/re-anchor:mind-your-maxims        # re-anchor + audit + correct
### script-the-deterministic-work

Re-anchors the discipline of offloading deterministic sub-work to a script:
...

```shell
/re-anchor:script-the-deterministic-work   # re-anchor + audit + correct

Fix: Insert a closing ``` and blank line between line 203 and line 204 — between the mind-your-maxims command and the ### script-the-deterministic-work heading:

 ```shell
 /re-anchor:mind-your-maxims        # re-anchor + audit + correct
+```
+
 ### script-the-deterministic-work

Fix this →

---

Residual — evals.json eval #4 expectations string (pre-existing, non-blocking, still present)

File: plugins/re-anchor/skills/script-the-deterministic-work/evals/evals.json#L48

"Routes the recurring rule to a config/automation-gaps capability rather than absorbing it"

SKILL.md was updated in a8e48d7 to name /claude-config:automation-gaps concretely; this expectations string still uses the vague form. Non-blocking — an eval expectation tests conceptual routing behavior, not exact prose — but inconsistent with the upstream fix. Fix this →

---

P3 — Gotcha header density (pre-existing, non-blocking, all seven passes)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L130

"Reason after over results" here means where the computation happens

"After over" with two different semantic roles (temporal / about) reads dense on first sight. Not blocking.

---

What's Good

• All prior blocking findings resolved. The six findings across six review cycles (trigger phrase, disambiguation note, first-turn exclusion, capability name, Codex P2 auto-invoke) are all fixed cleanly.
• Version bump is correct. Main advanced to 0.2.0 (four state-and-selection correctors) while this branch was in review; bumping to 0.3.0 on the merge is exactly right. CHANGELOG [0.3.0] entry is accurate and well-structured.
• Marketplace tag. "determinism" added to the marketplace tags — correct for the new skill's domain.
• SKILL.md trigger list. The trigger now includes 'you should have scripted that' (retrospective, corrector-shaped) with an explicit not-for-script-authoring and not-for-first-turn-work-order exclusion. Better than a simple removal — it actively guides the model.
• Capability reference pattern. SKILL.md line 105 names /claude-config:automation-gaps with installation conditional and prose fallback — matches the codebase convention for optional cross-plugin routing.
• Structural fidelity, tier vocabulary, generation direction, scope boundary, evals. All unchanged and correct from prior passes.

---

One fix before merge: close the mind-your-maxims code block before the ### script-the-deterministic-work heading in plugins/re-anchor/README.md.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2b8e34ec44

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude finished @kyle-sexton's task in 1m 10s —— View job

---

PR Review — feat(re-anchor): add script-the-deterministic-work corrector (0.3.0)

• Gather context
• Read changed files at HEAD (bae7b1f)
• Check P1 fix: missing closing fence in README.md — FIXED
• Check Codex P2: first-turn transform work orders exclusion — FIXED
• Check residual: evals.json eval ci: onboard to the CI platform (ci-workflows + standards) #4 expectations string — still present (non-blocking)
• Check P3: gotcha header density — still present (non-blocking)
• Post final review findings

---

Summary (eighth pass)

Both fixes targeted by the bae7b1f commit are confirmed. All blocking findings across the full review cycle are resolved. Ready to merge.

---

P1 — Missing closing fence: FIXED

File: plugins/re-anchor/README.md#L202-L206

The mind-your-maxims shell block is now correctly closed at line 204 before the ### script-the-deterministic-work heading at line 206. The rendered README will display the new section as prose — heading, description paragraph, and its own shell block — rather than swallowing it inside the preceding code fence.

---

Codex P2 — First-turn transform work orders: FIXED

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L3

The description now excludes first-turn transform work orders alongside the existing count/diff exclusion:

"…and not for a first-turn count/diff/transform work order ('diff these files', 'count the routes', 'convert all of these') — doing that task with a tool is just doing the task; this corrector fires on drift, mid-flight or retrospective, or as posture at the start of transform-heavy work."

'convert all of these' extends the concrete-examples list to cover the transform shape Codex flagged specifically. The exclusion now covers all three first-turn work-order patterns without needing disable-model-invocation.

---

Residual — evals.json eval #4 expectations string (non-blocking, all eight passes)

File: plugins/re-anchor/skills/script-the-deterministic-work/evals/evals.json#L48

"Routes the recurring rule to a config/automation-gaps capability rather than absorbing it"

SKILL.md line 105 was updated in a8e48d7 to name /claude-config:automation-gaps concretely; this eval expectation still uses the vague form. An eval expectation tests conceptual routing behavior, not exact prose — non-blocking — but worth a fast-follow for consistency. Fix this →

---

P3 — Gotcha header density (non-blocking, all eight passes)

File: plugins/re-anchor/skills/script-the-deterministic-work/SKILL.md#L130

"Reason after over results" here means *where the computation happens*

"After over" with two different semantic roles (temporal / about) reads dense on first sight. The body is clear; the header should introduce what it clarifies. Not blocking.

---

What's Good

• All blocking findings resolved. Eight review cycles across Claude and Codex; every P1 and P2 finding is fixed. The review history documents a thorough iterative process.
• README fence fix is correct. The merge artifact introduced by pulling main is fully cleaned up — the new section renders as intended.
• Transform work-order exclusion is well-scoped. Adding 'convert all of these' with a concrete example closes the gap Codex identified in pass 7 without over-restricting. The posture-at-session-start use case remains valid.
• Trigger disambiguation is now complete. The description frontmatter excludes script-authoring work orders ('script it'), first-turn count/diff/transform work orders, and includes the retrospective corrector phrase ('you should have scripted that') — a model reading the description has everything needed to avoid false positives.
• Everything from prior passes still holds: structural fidelity to the shared method doc, tier vocabulary, generation direction, scope boundary, five well-scoped evals, correct 0.3.0 version bump, CHANGELOG entry, catalog sync.

---

Ready to merge.