plan: run the AI-Native SDLC playbook — six stages, intent/spec/plan, no score (BREAKING) - #19
Merged
Conversation
…ec/plan artifacts, no score (BREAKING)
Nine phases become Plan, Design, Build, Test, Deploy, Maintain. Each stage
commits one artifact the next reads, all in the plan folder: intent.md,
spec.md (+ audit.md from the verifier gate), plan.md (files, order, risks,
proof, verification, parallelization), test-plan.md, review.md, and a new
Draft intent proposed from SEV1/SEV2 or recurring incidents.
skills/plan: SKILL.md router (lifecycle table, workspace, Step 0 with
`intent {name}` mode and a schema-1 -> schema-2 migration map, stage
dispatch, staleness cascade); stages/stage-1..6; templates/intent, spec,
plan, review. Old phase bodies carried verbatim into the stages; the design
gate keeps <gate_expression>, <revision_feedback>, CANARY, EVIDENCE
VALIDATION, PLAN LINT RULES and the rubric-free pass byte-identical. Removed:
the 8-dimension score, /40, bands, score_history, dimension_scores, and the
score clause in the waiver (now infra_gaps, canary_found, evidence_demotions).
Deploy: fresh-subagent diff-versus-plan check (unplanned/untouched files,
intent constraints vs the code) that must be acknowledged in plan.md before
release; the skill never runs a release command and records authorized_by.
Maintain: troubleshoot step 5 proposes a Draft intent for linked incidents.
status.json keeps `current_phase` and `phases` keys (hooks unchanged); stage
entries carry started/completed timestamps; plan-status prints intent->spec,
spec->plan, plan->release elapsed and reads either schema.
plan-auditor: scoring layer removed (542 lines from 603), verdicts per
criterion only; <forbidden_inputs> and <verdict_schema> intact.
Tests: PH5-* re-pointed to stages/stage-2-design.md; PH5-010 judge floor is
now the eight review dimensions; PH5-051 inverted to assert no scoring
vocabulary survives; PH5-060 checks the new waiver terms; sweeps include
skills/*/stages. Docs, help, GUIDE, plugin and marketplace descriptions,
interop rows, mutation-guard paths updated.
Verified: run-all.sh all 8 layers green; claude plugin validate passes on the
plugin and the marketplace. Not verified: a live end-to-end plan run on the
new skill (a --plugin-dir probe resolved the installed 7.1.0 copy instead).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YDwLDqUNM5U8i23bDDWFFp
A resume of the real plugin-hardening-v5 folder (schema 1) against the working-tree skill migrated correctly but hit three ambiguities: a stage with one complete and one unstarted old phase, a non-standard status (`v17_complete`), and an old key the map does not name (`review`). Step 0 now states an ordered status rule (suffix "complete" counts; all complete -> complete; any complete/in_progress -> in_progress; else not_started), keeps the whole old phases object under `phases_schema1` with a `schema1_migration` block, and tells the resume summary to say when a later stage finished before an earlier one. Probes run from a scratch repo with CLAUDE_SKILL_DIR pointed at the working tree: (a) `intent json-flag from idea: ...` produced intent.md with every template section and an Accepted header, manifest.md, research/, and a schema-2 status.json with plan started/completed/accepted_by, then stopped; (b) resume of the schema-1 plan wrote schema 2 with the six stages, left brainstorm.md/approach.md/manifest.md byte-identical (md5), and stopped at the summary. Both status.json files parse with node. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YDwLDqUNM5U8i23bDDWFFp
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Restructures /deepgrade:plan onto Anthropic's six-stage playbook. Each stage commits one artifact in the plan folder: intent.md, spec.md + audit.md, plan.md, test-plan.md, review.md, and a Draft intent from incidents. The verifier gate is kept byte-for-byte; the numeric score is removed. Deploy adds a diff-versus-plan check and never runs a release. status.json moves to schema 2 with a migration map for old plans; hooks are unchanged.
Decisions locked with Kyle: 1A 2B 3B 4B (score removed, registry kept) 5A.
All 8 test layers green; validate passes. Not yet verified: a live end-to-end plan run on the new skill.
🤖 Generated with Claude Code
https://claude.ai/code/session_01YDwLDqUNM5U8i23bDDWFFp