skill-evolution: collapse three-phase model to single workflow

[ramakrishnap-nv]

Summary

Merges Phase 1 (Learning), Phase 2 (Inference), and Phase 3 (Reflection) into a single Workflow section in skills/skill-evolution/SKILL.md. The phases were really one process gated by whether a score was available — the split added ceremony without changing the steps.

Three sub-changes bundled:

• One workflow. Solve → notice trigger → try to score → distill, place, propose. Confidence is now a scored: yes/no field on the proposal, not a separate pipeline.
• No hard iteration cap. The "max 3 iterations" sandbox limit was unenforceable in interactive sessions. Replaced with "iterate until it scores or drop it."
• Compressed proposal format. From 7 fields (Skill, Type, Phase, Section, Trigger, Score, Change) to 4 (Target, Trigger, Scored, Diff). The diff itself carries Type and Section.

Reflection's "recurrence as evidence" rule is preserved as step 6 of the workflow. Distillation, Placement rule, Scoring criteria, and Provenance tagging are unchanged.

Net: 75 lines removed, 18 added.

Testing

• ./ci/utils/validate_skills.sh passes

Docs

• Skill content only.

[coderabbitai]

📝 Walkthrough

Walkthrough

Rewrites the skill-evolution workflow from three phases into a single solve-first, then evaluate-and-refine flow; removes the sandboxed learning phase; changes proposal schema and recurrence handling; and simplifies the compliance/provenance checklist. (Lines changed: +20/-76)

Changes

Cohort / File(s)

Summary

Skill Evolution Documentation skills/skill-evolution/SKILL.md

Replaced three‑phase lifecycle (learning/inference/reflection) with a single solve‑then‑score workflow; removed sandboxed learning loop; updated recurrence handling to upgrade repeated unscored insights via Trigger; overhauled proposal fields to Target, Trigger, Scored, Removal, Diff; deleted separate inference/offline reflection sections; simplified compliance checklist to require Scored: (validation details or no). Net: +20/-76.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title directly summarizes the main change: collapsing a three-phase model into a single workflow, which matches the core refactoring described in the raw summary.
Description check	✅ Passed	The description clearly explains the consolidation of three phases into one workflow, details the sub-changes (unified workflow, iteration model, compressed format), and is directly related to the changeset.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Review rate limit: 7/10 reviews remaining, refill in 16 minutes and 56 seconds.}

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

🧹 Nitpick comments (1)

skills/skill-evolution/SKILL.md (1)

27-34: Consider clarifying the distinction between steps 3 and 4.

Steps 3 and 4 work together but could be slightly clearer:

• Step 3 applies when ground truth exists but the learning doesn't validate → iterate or drop
• Step 4 applies when ground truth doesn't exist → proceed with scored: no

The parenthetical "(no test, no comparable answer, no solver run)" in step 4 does clarify this, but the phrase "If you cannot score it at all" could briefly be misread as "if you tried and failed" rather than "if no scoring method is available."

The current wording is functional and the parenthetical resolves the ambiguity, but you might consider rewording step 4's opening to: "If no ground truth is available to score against..." for extra clarity.

📝 Optional rewording for clarity

-4. **If you cannot score it at all** (no test, no comparable answer, no solver run), proceed with `scored: no`. This is normal during inference-style interactions — the proposal is still useful, just lower-confidence.+4. **If no ground truth is available** (no test, no comparable answer, no solver run), proceed with `scored: no`. This is normal during inference-style interactions — the proposal is still useful, just lower-confidence.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.
In `@skills/skill-evolution/SKILL.md` around lines 27 - 34, Reword step 4 to
remove ambiguity by explicitly contrasting it with step 3: change the opening
sentence from "If you cannot score it at all" to something like "If no ground
truth is available to score against" and keep the existing parenthetical "(no
test, no comparable answer, no solver run)"; ensure step 3 remains described as
the path when ground truth exists and scoring fails (iterate or drop), and
update the surrounding sentences so Steps 3 and 4 clearly read as complementary
branches rather than sequential attempts.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@skills/skill-evolution/SKILL.md`:
- Around line 27-34: Reword step 4 to remove ambiguity by explicitly contrasting
it with step 3: change the opening sentence from "If you cannot score it at all"
to something like "If no ground truth is available to score against" and keep
the existing parenthetical "(no test, no comparable answer, no solver run)";
ensure step 3 remains described as the path when ground truth exists and scoring
fails (iterate or drop), and update the surrounding sentences so Steps 3 and 4
clearly read as complementary branches rather than sequential attempts.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 93ff09e7-dd51-4d54-bad6-0d8d62f380d4

📥 Commits

Reviewing files that changed from the base of the PR and between 9c48fba and 7e61345.

📒 Files selected for processing (1)

• skills/skill-evolution/SKILL.md

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@skills/skill-evolution/SKILL.md`:
- Around line 85-86: The document contains a contradiction between the proposal
schema's "Removal: no | yes" field and the Security rules that forbid removing
existing content; resolve by either (A) removing the "Removal" field and any
associated logic that requires explicit user confirmation (search for the
"Removal:" entry in SKILL.md and delete it and its references), or (B) update
the Security rules to allow removals conditionally when "Removal: yes" with
explicit user confirmation (locate the security section that says "A proposal
must NOT: Remove existing content" and change it to permit removals when the
proposal's Removal field is set to yes and user confirmation is recorded), then
ensure all cross-references and examples in SKILL.md reflect the chosen
approach.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: f4dc75d9-3cb6-449c-bc5f-7ab1f7b1e27a

📥 Commits

Reviewing files that changed from the base of the PR and between 7e61345 and 23dccb7.

📒 Files selected for processing (1)

• skills/skill-evolution/SKILL.md