32 KiB
Memory v2 write side (phase 4a) Implementation Plan
For agentic workers: REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (
- [ ]) syntax for tracking.
Goal: lib/memory.sh becomes the only writer of .claude/memory registries, one status parser serves memory.sh and reconcile, this repo's registries move to v2 (no ## Index), every citer writes through the lib.
Architecture: One sourceable bash lib on the lib/project-rules.sh skeleton (-C <dir>, verbs, rc 2 + named reason). The awk status parser lives once in _MEM_AWK and is shared by index, upgrade and lib/reconcile.sh. Writers serialize through a noclobber lock file in the git common dir. memory-commit.sh runs check --staged as the chokepoint.
Tech Stack: bash 5, POSIX awk (no gawk-only features, no {n} intervals), git. Tests via make test.
Spec: docs/superpowers/specs/2026-10-27-memory-v2-write-design.md (user-approved 2026-10-27). Contract: .claude/tasks/contracts/2026-10-27-memory-v2-write-2030.md.
Reference implementation (proven before this plan was written): docs/superpowers/plans/memory-v2-write/ holds memory.sh, memory.test.sh and blockers-v2-status.md. They were run in a mock repo: suite PASS=72 FAIL=0 five runs in a row, shellcheck clean, no line over 80 chars; upgrade on a copy of this repo's live registries is oracle-green, check-green, idempotent, and its diff passes check --staged. Task 1 copies them; do not rewrite them.
Global Constraints
- Branch
feature/memory-v2-write(already started, spec atcdc0a47). Never switch, never merge, never push by hand (hooks push). - Our code: functions ≤ 25 logic lines (an embedded awk program counts as one statement), ≤ 5 locals, ≤ 5 params, lines ≤ 80 chars in
lib/*.sh. - No
# shellcheck disableanywhere in the diff (floor guard, LRN-234).# shellcheck source=/dev/nulldirectives are allowed. /usr/bin/greppinned in libs (MEM_GREP, LRN-074); nogrep -qreading a pipe under pipefail (LRN-189); nosed -i,stat -c,realpath -m,grep -P(portability census, BDR-110).- No recursive delete through a variable except the existing
trap 'rm -rf "$FIX"' EXITmktemp pattern. - Hermetic tests run through
make test [suite=…]only. - Census-cited strings stay byte-identical: every
CLAUDE.md "Memory registries" (Always English, always caveman)citation, the CLAUDE.global.md heading## Memory registries (\.claude/memory/`)and the bold labelAlways English, always caveman` (doctrine-citers, LRN-166). Single-line fixed strings: do not rewrap a line that holds one. - CLAUDE.global.md line count must not exceed 333 after the edit.
- Worked examples use fictional 9xx IDs (LRN-073).
- Entry text written to registries is English caveman (BDR-009).
Deviations from the spec (found while proving the reference)
- Lock = a file created under bash noclobber, not a
mkdirdir: uutilsmkdir0.10 on this host is not atomic (5 concurrent adds lost 2-4 entries). Same path (<git-common-dir>/memory.lock), same 30 s steal. journaltakes no--dateflag:MEMORY_DATEalready pins the date for tests; YAGNI.upgradewrites an Index status into the body only when the body has none, or when the Index says superseded/deprecated and the body does not. Otherwise the body wins: the Index is the stale copy (BDR-012: Index accepted, body shipped). Status words rank in_MEM_WORDSorder withopenlast, measured on the live registries: identical verdicts to today's reconcile on every blocker except BLK-032..034.
Review Focus
- Concurrent writers. Two sessions capitalizing at once must get two IDs; a lock that is not atomic loses entries silently. Found while proving the reference:
mkdirfrom uutils coreutils 0.10 (this host) is NOT atomic, two concurrentmkdirboth succeed; the lock uses bash noclobber (open(O_EXCL)). T6e/T6f are the flip test (they were red with the mkdir lock). - Status words in prose. A Solution line saying "fail-open" or "
data:OPEN" must never reopen a resolved blocker; "accepted. Behavioral check OPEN" must read accepted. Covered by T10b/c/e and run-reconcile T8. - Index-only facts lost at upgrade. 13 BDR statuses, 2 BDR + 2 BLK + 12 LRN supersessions and ~130 dates live only in Index rows today; a stale Index status (BDR-012 says accepted, body says shipped) must NOT overwrite the body. The upgrade oracle refuses on any loss; T9e pins the stale case.
- v1 projects keep working. 13 other repos on this machine run this lib through
~/.claude/lib; a v1 registry is written leniently,checknever fails on it,--stagedignores it, memory-commit keeps committing it. T7e/f and the memory-commit v1 test pin this. - The chokepoint must not block the upgrade commit itself. Removing the Index (lines before the first entry) and inserting Date/Status lines are not "removed lines inside an entry". Proven on the live copy; Task 7 commits it through memory-commit.sh.
Task 1: lib/memory.sh, its suite and the parser fixture
Files:
- Create:
lib/memory.sh(copy ofdocs/superpowers/plans/memory-v2-write/memory.sh) - Create:
lib/tests/memory.test.sh(copy ofdocs/superpowers/plans/memory-v2-write/memory.test.sh) - Create:
lib/tests/fixtures/blockers-v2-status.md(copy ofdocs/superpowers/plans/memory-v2-write/blockers-v2-status.md)
Interfaces:
-
Produces (sourceable, used by Tasks 2-7):
memory_entry_status <file> <ID>(winning candidate line, rc 1 if none),memory_status_word <line>(status word or first 40 chars),memory_check [--staged] [<dir>](rc 2 on a culprit),memory_add,memory_status,memory_journal,memory_index,memory_upgrade. CLI:bash lib/memory.sh [-C dir] add|status|journal|index|check|upgrade …. Env:MEMORY_DATE,MEM_GREP,MEM_LOCK_TRIES. -
Step 1: Copy the three files
cp docs/superpowers/plans/memory-v2-write/memory.sh lib/memory.sh
cp docs/superpowers/plans/memory-v2-write/memory.test.sh lib/tests/memory.test.sh
cp docs/superpowers/plans/memory-v2-write/blockers-v2-status.md lib/tests/fixtures/blockers-v2-status.md
chmod +x lib/memory.sh lib/tests/memory.test.sh
- Step 2: Run the suite and the static checks
Run: make test suite=lib/tests/memory.test.sh → expected last lines PASS=72 FAIL=0 and all suites green. Run it three times (the race test T6e must be stable).
Run: shellcheck lib/memory.sh lib/tests/memory.test.sh && awk 'length > 80' lib/memory.sh → no output after shellcheck.
Run: make test suite=lib/tests/portability-census.test.sh → green.
- Step 3: Commit
git add lib/memory.sh lib/tests/memory.test.sh lib/tests/fixtures/blockers-v2-status.md
git commit -m "feat(memory): lib/memory.sh single registry writer — add/status/journal/index/check/upgrade, shared status parser, noclobber lock"
Task 2: reconcile reads status through the shared parser
Files:
- Modify:
lib/reconcile.sh:57-67(reconcile_blk_current_status) and its header comment:8-12 - Modify:
lib/tests/run-reconcile.sh(new T8 block before the final summary line) - Modify:
skills/reconcile/SKILL.md:25
Interfaces:
-
Consumes:
memory_entry_statusfrom Task 1. -
Produces:
reconcile_blk_current_status <file> <ID>unchanged signature, now prints the winning candidate line (empty when none). -
Step 1: Write the failing test — insert before the
echo; echo "================ $pass GREENline oflib/tests/run-reconcile.sh:
echo; echo "=== T8 shared parser — header [tag] + labelled lines, prose never (BLK-032..034) ==="
v2="$FIX/blockers-v2-status.md"
open_v2=$(reconcile_blk_open "$v2" | cut -f1 | sort | tr '\n' ' ')
if [ "$open_v2" = "BLK-041 BLK-042 " ]; then ok "T8a open = {041,042}: 032..034, 008, 040 read resolved/accepted"; else no "T8a open = [$open_v2], expected {041,042}"; fi
case "$(reconcile_blk_current_status "$v2" BLK-034)" in
*'[resolved]'*) ok "T8b BLK-034 status = its own header tag";;
*) no "T8b BLK-034 read from prose (fail-open), not the header tag";;
esac
- Step 2: Run it, expect red
Run: make test suite=lib/tests/run-reconcile.sh → RED ✗ T8a open = [BLK-032 BLK-033 BLK-034 BLK-040 BLK-041 BLK-042 ]… (exact list may differ; it must be red).
- Step 3: Delegate the parser — in
lib/reconcile.sh, after theRECONCILE_GREP=line add:
# the status parser is shared with lib/memory.sh (BDR-125 phase 4a)
# shellcheck source=/dev/null
. "$(dirname "${BASH_SOURCE[0]}")/memory.sh"
and replace the whole reconcile_blk_current_status function and the two comment lines above it with:
# --- registry status: LAST candidate wins (the BLK-008 trap A fell into) ---
# $1 blockers file, $2 id. Echoes the winning status line: the entry's own
# header `[tag]` or a labelled body line (lib/memory.sh); a status word in
# prose is never read (the BLK-032..034 false-open). Empty when none.
reconcile_blk_current_status() {
memory_entry_status "$1" "$2"
}
In the header comment, insert one line right after line 12 (# preaches — the recursive-coherence test (run-reconcile.sh T1) reds if this is broken.):
# v2 registries carry no Index at all (lib/memory.sh); v1 ones still may.
- Step 4: Run, expect green
Run: make test suite=lib/tests/run-reconcile.sh → T8a, T8b green, T1-T7 still green, 0 RED.
Run: make test suite=lib/tests/run-review-guards.sh → green (G4: run-reconcile must not read .claude/memory; T8 reads the fixture).
-
Step 5: Skill line —
skills/reconcile/SKILL.md:25, replace(`reconcile_oracle_*`, `reconcile_blk_current_status` = last block wins)by(`reconcile_oracle_*`, `reconcile_blk_current_status` = last candidate wins: header `[tag]` or a labelled status line, via `lib/memory.sh`) -
Step 6: Commit
git add lib/reconcile.sh lib/tests/run-reconcile.sh skills/reconcile/SKILL.md
git commit -m "fix(reconcile): blocker status through the shared memory.sh parser — BLK-032..034 no longer false-open"
Task 3: memory-commit.sh refuses a staged in-place edit (rc 6)
Files:
- Modify:
lib/memory-commit.sh(header usage comment,commit_memorybetweengit addandgit diff --cached --quiet) - Modify:
lib/tests/run-deterministic.sh(two new tests before the summary) - Modify:
lib/capitalize-commit.md(rc table)
Interfaces:
-
Consumes:
bash lib/memory.sh check --staged .claude/memory. -
Produces:
memory-commit.sh commitrc 6 = check refused, nothing committed. -
Step 1: Write the failing tests — read
lib/tests/run-deterministic.shfirst (helpersnew_repo,ok,ko,HELPER); append before its final summary block:
echo "== check --staged chokepoint (v2 registry, phase 4a)"
d="$(new_repo)"
printf -- '---\nformat: 2\ntype: decisions_registry\nentry_prefix: BDR\n---\n\n## BDR-001 — kept (2026-10-27)\n- **Status**: accepted\n' \
> "$d/.claude/memory/decisions.md"
git -C "$d" add -A && git -C "$d" commit -qm v2
head_before=$(git -C "$d" rev-parse HEAD)
printf -- '---\nformat: 2\ntype: decisions_registry\nentry_prefix: BDR\n---\n\n## BDR-001 — rewritten (2026-10-27)\n- **Status**: accepted\n' \
> "$d/.claude/memory/decisions.md"
out="$(cd "$d" && bash "$HELPER" commit "chore(memory): bad" 2>/dev/null)"; rc=$?
if [ "$rc" -eq 6 ] && [ -z "$out" ] && [ "$(git -C "$d" rev-parse HEAD)" = "$head_before" ]; then ok "in-place edit of a v2 entry → rc 6, nothing committed"; else ko "rc=$rc out=[$out] — in-place edit was committed or wrong rc"; fi
git -C "$d" checkout -q HEAD -- .claude/memory/decisions.md
printf '\n## BDR-002 — added (2026-10-27)\nbody\n' >> "$d/.claude/memory/decisions.md"
out="$(cd "$d" && bash "$HELPER" commit "chore(memory): BDR-002" 2>/dev/null)"; rc=$?
if [ "$rc" -eq 0 ] && [ -n "$out" ]; then ok "append to a v2 registry → committed"; else ko "append refused rc=$rc"; fi
(The existing tests write baseline registries with no frontmatter: v1, check skips them, they must stay green unchanged.)
- Step 2: Run, expect red
Run: make test suite=lib/tests/run-deterministic.sh → the rc 6 test fails (today rc 0, the edit is committed).
- Step 3: Implement — in
lib/memory-commit.sh:- under
MC_PATHS=(...)add:
- under
MC_MEMORY_SH="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)/memory.sh"
- in
commit_memory, right aftergit add -- "${changed[@]}", insert:
# Chokepoint (BDR-125 phase 4a): a staged v2 registry may only grow. An
# in-place rewrite of an existing entry is curation: /prune-memory, then
# /commit-change. v1 registries are skipped by the check.
if [ -d .claude/memory ] \
&& ! bash "$MC_MEMORY_SH" check --staged .claude/memory; then
echo "memory-commit: REFUSED — memory.sh check --staged failed (above);" \
"NOTHING committed, the scoped paths stay staged." >&2
return 6
fi
-
header usage line becomes
# memory-commit.sh commit "<message>" # surgical commit; exit 0 ok/no-op, 3 unsafe state, 5 commit rejected, 6 check --staged refused -
Step 4: Run, expect green
Run: make test suite=lib/tests/run-deterministic.sh → all green including the two new ones.
- Step 5: Document rc 5 and rc 6 — in
lib/capitalize-commit.mdstep 3 "Report by (rc, mem_hash)", after therc 3bullet add:
- rc 5 → `git commit` rejected (hook, protected branch,
signing); nothing committed, surface stderr.
- rc 6 → `memory.sh check --staged` refused: a staged
v2 registry rewrites an existing entry line
(only appends pass). Curation goes through
/prune-memory then /commit-change; surface
the culprit lines, never bypass the helper.
- Step 6: Commit
git add lib/memory-commit.sh lib/tests/run-deterministic.sh lib/capitalize-commit.md
git commit -m "feat(memory): memory-commit refuses a staged in-place edit of a v2 registry (rc 6); rc 5/6 documented"
Task 4: v2 templates and the registry v1 banner line
Files:
-
Modify:
templates/memory/{decisions,learnings,blockers,evals}.md -
Modify:
hooks/session-start.sh(after the CLAUDE.global.md line-count guard, before# Version check) -
Modify:
lib/tests/memory.test.sh(T13, before the finalecho "PASS=…") -
Modify:
agents/onboarder.md:126(template description) -
Step 1: Write the failing test — insert before
echo "PASS=$pass FAIL=$fail"inlib/tests/memory.test.sh:
# T13 templates are born v2; the session banner names a v1 project only
for t in decisions learnings blockers evals; do
check "T13-template-$t-v2" "$(sed -n 2p "$ROOT/templates/memory/$t.md")" 'format: 2'
check "T13b-template-$t-no-index" "$(grep -c '^## Index$' "$ROOT/templates/memory/$t.md")" 0
done
banner() { (cd "$1" && SESSION_START_OFFLINE=1 bash "$ROOT/hooks/session-start.sh" \
</dev/null 2>/dev/null); }
check T13c-banner-v1 "$(banner "$FIX/v1-proj" | grep -c 'registry v1: 1 file')" 1
check T13d-banner-v2 "$(banner "$FIX" | grep -c 'registry v1')" 0
and, just above it, the v1 fixture project:
mkdir -p "$FIX/v1-proj/.claude/memory"
printf '# Decisions\n\n## BDR-001 - legacy\n' > "$FIX/v1-proj/.claude/memory/decisions.md"
-
Step 2: Run, expect red —
make test suite=lib/tests/memory.test.sh→ T13 checks fail. -
Step 3: Templates — in each of the four templates: insert
format: 2as line 2 (right after the opening---); delete the## Indexheading, the blank line after it, the two table lines and the blank line after them; replace the<!-- Append entries below. Template: … -->comment by this shape (per file, adapt the kind, title word and body labels already in that template; the heading line is indented two spaces so no line starts with##):
<!-- Written only by `bash ~/.claude/lib/memory.sh add blocker --title "<friction>"
--body-file <file> [--status open]` (it allocates the ID). Entry shape:
## BLK-NNN — <friction> [open] (YYYY-MM-DD)
- **Friction**: <what was blocked>
- **Real cause**: <root cause>
- **Solution**: <workaround or fix>
Later status: `bash ~/.claude/lib/memory.sh status BLK-NNN "resolved <hash>"`.
-->
Kinds and labels: decisions → decision, <title>, [accepted], labels Decision / Why / Rejected alternatives / Reference; learnings → learning, <abstract pattern>, no status, labels Pattern / Context / Future application; evals → eval, <output>, [keep], labels Output / Method / Anomalies / Action. journal.md template unchanged.
- Step 4: Banner — in
hooks/session-start.sh, after thefithat closes the CLAUDE.global.md line-count guard, insert:
# Registry v1 (BDR-125 phase 4a): a project registry without `format: 2`.
if [ -n "$REPO_DIR" ] && [ -d .claude/memory ]; then
_v1=$(bash "$REPO_DIR/lib/memory.sh" check .claude/memory 2>&1 >/dev/null \
| grep -c '^memory: v1: ' || true)
if [ "${_v1:-0}" -gt 0 ]; then
_v1_line="registry v1: ${_v1} file(s), see memory.sh upgrade"
printf "│ ⚠️ %-44s│\n" "${_v1_line:0:44}"
fi
unset _v1 _v1_line
fi
-
Step 5: Onboarder line —
agents/onboarder.md:126, replace(YAML schema + empty index + inline template comment)by(YAML schema with \format: 2` + inline template comment; no Index, entries go through `lib/memory.sh`)`. -
Step 6: Run, expect green —
make test suite=lib/tests/memory.test.sh→FAIL=0;make test suite=lib/tests/push-guard.test.sh→ green (banner unchanged for this repo, which is v1 until Task 7: if push-guard counts banner lines exactly, read its T44-T46 and confirm none breaks; the new line appears only in a project with a v1 registry). -
Step 7: Commit
git add templates/memory hooks/session-start.sh lib/tests/memory.test.sh agents/onboarder.md
git commit -m "feat(memory): templates born v2 (format: 2, no Index); session banner names a v1 registry"
Task 5: flows write through the lib (capitalize, CAPITALIZE steps, analyzer, doctrine)
Files (one edit each, exact text below):
skills/capitalize/SKILL.mdSTEP 2 (:143), STEP 4 items 1-4 (:247-255), STEP 5 (:267-273)skills/feat/SKILL.md:248skills/bugfix/SKILL.md:275skills/hotfix/SKILL.md:234(+ the two proposal bullets:228-229unchanged)skills/ship-feature/SKILL.md:273-274skills/init-project/SKILL.md:301-303agents/commit-changer.md:220-223skills/onboard/SKILL.mdS7 (:1181-1183)lib/analyze-before-plan.md:42-43and:100-102agents/analyzer.md:98-99CLAUDE.global.md§ Memory registries (two sentences)
The one canonical write instruction (W) used below:
W = Write each approved entry's caveman body to a scratch file, then
bash "$HOME/.claude/lib/memory.sh" add <decision|learning|blocker|eval> --title "<title>" --body-file <file> [--status <status>](one call per entry; it allocates the ID and prints it; never Edit or append to a registry, there is no Index to update). A status change on an existing entry:bash "$HOME/.claude/lib/memory.sh" status <ID> "<text>". The journal line:bash "$HOME/.claude/lib/memory.sh" journal "<line>".
- Step 1: capitalize
:143(Index + bodies)→(headings + bodies).- STEP 4 items 1-4 (from
1. Read the target registry file.through the Index-row item endingmatch that file's header row).) become:
1. Body to a scratch file, then `bash "$HOME/.claude/lib/memory.sh" add
<decision|learning|blocker|eval> --title "<title>" --body-file <file>
[--status <status>] [--supersedes <ID>]`: it allocates the next ID from
the headings under a lock, appends at the end, prints the ID. Never Edit
or append to a registry by hand; there is no Index to update.
2. A status change on an existing entry (a blocker now resolved):
`bash "$HOME/.claude/lib/memory.sh" status <ID> "<text>"`.
3. A refusal (rc 2) names its reason (title chars, empty body, a blocker
body without one backtick-quoted error): fix the input, re-run.
Old item 5 (caveman) stays, renumbered 4.
-
STEP 5: replace the two bullets
- Heading exists → append a bullet./- Missing → create …by- \bash "$HOME/.claude/lib/memory.sh" journal ""` (one call per bullet; it creates today's heading when missing).` -
Step 2: feat
:248(Always append a 1-line entry to today's heading in \.claude/memory/journal.md`.`) → two lines, the Language rule line after them untouched:
Approved entries: caveman body to a scratch file, then `bash "$HOME/.claude/lib/memory.sh" add <decision|learning> --title "<title>" --body-file <file>` (one call each; it allocates the ID; never Edit a registry).
Always append a 1-line entry: `bash "$HOME/.claude/lib/memory.sh" journal "<line>"`.
-
Step 3: bugfix
:275→4. Write approved entries with \bash "$HOME/.claude/lib/memory.sh" add --title … --body-file --status resolved` (one call each, it allocates the ID; an existing blocker now fixed: `memory.sh status "resolved "`). Journal line: `bash "$HOME/.claude/lib/memory.sh" journal ""`.` -
Step 4: hotfix
:234→Always append a 1-line entry with \bash "$HOME/.claude/lib/memory.sh" journal ""` (even trivial hotfix — journal is timeline, not signal). A proposed BLK/LRN, once approved: `bash "$HOME/.claude/lib/memory.sh" add <blocker|learning> --title … --body-file `.` -
Step 5: ship-feature
:273-274→
4. Write approved entries with `bash "$HOME/.claude/lib/memory.sh" add <kind> --title "<title>" --body-file <file> [--status <s>]`, one call each (it allocates the ID; no Index to update).
5. Journal line: `bash "$HOME/.claude/lib/memory.sh" journal "<line>"`.
-
Step 6: init-project
:301-303→3. Write approved entries to the registries bootstrapped at STEP 5 (in the root commit) with \bash "$HOME/.claude/lib/memory.sh" add --title … --body-file `, one call each; journal line with `bash "$HOME/.claude/lib/memory.sh" journal ""`.` -
Step 7: commit-changer
:220-223(item 2 of Phase 4) →
2. Write each resolved entry with `bash "$HOME/.claude/lib/memory.sh" add
<decision|learning|blocker|eval> --title "<title>" --body-file <file>`
(one call each; it allocates the ID, no Index to update), then one
journal line for the batch: `bash "$HOME/.claude/lib/memory.sh" journal
"<summary>"`.
-
Step 8: onboard S7 — replace
append to that project's \.claude/memory/journal.md` under today's date:byappend to that project's journal with `bash "$HOME/.claude/lib/memory.sh" -C journal`:` (the quoted line text after it unchanged). -
Step 9: analyze-before-plan + analyzer
lib/analyze-before-plan.md:42-43:Do NOT read the \## Index` table: it drifts (entries land in the body, the manual Index update lapses — measured 32-40% missing on a mature repo); headings cannot.→Do NOT read a `## Index` table (v1 registries may still carry one; it drifted 32-40% on a mature repo; v2 has none); headings cannot drift.`:100-102:No write, no Index update, no memory mutation.→No write, no memory mutation.andIndex backfill, if ever wanted, is \/prune-memory` passe D — never this snippet.→Registry writes go through `lib/memory.sh`, never this snippet.`agents/analyzer.md:98-99:Index backfill or new entries are never your job→new entries or status lines are never your job (writes go through lib/memory.sh).
-
Step 10: CLAUDE.global.md (line-neutral):
Append-only: never rewrite past entries; curation (merge, supersede,\ncompress) only via \/prune-memory`.→Append-only, written only through `bash ~/.claude/lib/memory.sh` (add,\nstatus, journal), never by Edit; curation only via `/prune-memory`.`exact. Pattern \[thing] [action] [reason]. [next step].` Registries load\nevery session; caveman cuts ~40% of the tokens with no substance lost.→exact. Pattern `[thing] [action] [reason]. [next step].` Planning flows\nread the registries; caveman cuts ~40% of every read, no substance lost.`
-
Step 11: Verify Run:
grep -rn 'Index' skills/capitalize/SKILL.md skills/feat/SKILL.md skills/bugfix/SKILL.md skills/hotfix/SKILL.md skills/ship-feature/SKILL.md skills/init-project/SKILL.md agents/commit-changer.md→ no line telling a writer to update an Index. Run:wc -l < CLAUDE.global.md→ ≤ 333. Run:make test suite=lib/tests/doctrine-citers.test.shandmake test suite=lib/tests/effort-routing.test.sh→ green (no rewrap of a census string; everyroute(phase="apply") … memory-commitline untouched). -
Step 12: Commit
git add skills/capitalize/SKILL.md skills/feat/SKILL.md skills/bugfix/SKILL.md skills/hotfix/SKILL.md skills/ship-feature/SKILL.md skills/init-project/SKILL.md agents/commit-changer.md skills/onboard/SKILL.md lib/analyze-before-plan.md agents/analyzer.md CLAUDE.global.md
git commit -m "docs(memory): every writer goes through lib/memory.sh; Index steps removed; doctrine names the single writer"
Task 6: /prune-memory on v2 (no § D, status through the lib, check as verify)
Files:
-
Modify:
skills/prune-memory/SKILL.md(:83-85,:111-115,:130,:141-143,:153-158,:177, STEP 4 block:190-206, report:274, failure row:305,:322-323) -
Modify:
skills/prune-memory/tests/run-deterministic.sh(RED-6 rewritten) -
Create:
skills/prune-memory/tests/fixtures/red6-check/.claude/memory/decisions.md -
Delete:
skills/prune-memory/tests/fixtures/red6-orphan/.claude/memory/decisions.md(git rm) -
Modify:
skills/prune-memory/test-prompts.json(id 1expected) -
Step 1: Write the failing test — replace the whole RED-6 block (from
# ---- RED-6:through its closingfi) inskills/prune-memory/tests/run-deterministic.shby:
# ---- RED-6: STEP 4 verify runs memory.sh check (duplicate ID caught) --------
# v2 registries carry no Index; the structural verify is `memory.sh check`
# (heading shape, unique IDs). HOME is pinned to a sandbox whose .claude/lib
# is this repo's lib, so the run never depends on the machine's ~/.claude.
S6="$SANDBOX/red6"; mkdir -p "$S6/.claude/memory" "$S6/home/.claude"
ln -s "$HERE/../../../lib" "$S6/home/.claude/lib"
cp "$HERE/fixtures/red6-check/.claude/memory/decisions.md" \
"$S6/.claude/memory/decisions.md"
extract_block "STEP 4" > "$S6/step4.sh"
out="$( cd "$S6" && HOME="$S6/home" bash step4.sh 2>&1 )"
if printf '%s\n' "$out" | grep -qE 'duplicate BDR-009'; then
green 6 "verify runs memory.sh check and names the duplicate BDR-009"
else
red 6 "verify misses the duplicate BDR-009 (no memory.sh check in STEP 4)"
fi
Create skills/prune-memory/tests/fixtures/red6-check/.claude/memory/decisions.md:
---
format: 2
type: decisions_registry
entry_prefix: BDR
---
# Decisions registry (BDR)
## BDR-009 — first (2026-06-01)
- **Status**: accepted
## BDR-009 — duplicate, the trap (2026-06-02)
- **Status**: accepted
git rm -q skills/prune-memory/tests/fixtures/red6-orphan/.claude/memory/decisions.md. In the suite header comment, RED-2/RED-6 execute the REAL bash blocks stays true.
-
Step 2: Run, expect red —
make test suite=skills/prune-memory/tests/run-deterministic.sh→RED-6: RED. -
Step 3: SKILL.md edits
- § A bullet
:83-85:Decisions whose body contains "superseded by <ID>" but Index row still says \accepted` → propose Index row fix to `superseded by `.→Decisions whose body names a successor ("superseded by ", "Supersedes [[]]" on the newer one) but whose current status (`bash "$HOME/.claude/lib/memory.sh" index decision`) still reads `accepted` → propose `memory.sh status "superseded by "`.` - Delete § D (
### D. Index driftand its two bullets,:111-115). - STEP 2 example:
→ fix Index: status = "superseded by BDR-903"→→ memory.sh status BDR-902 "superseded by BDR-903"; delete the[D. Index drift]+(none)lines;Approve per category? (all / a / b / c / d / edit <ID> / skip)→(all / a / b / c / edit <ID> / skip). - STEP 3 list becomes:
- § A bullet
1. **Status flag updates** — `bash "$HOME/.claude/lib/memory.sh" status <ID>
"<text>"` appends a dated status line; the body above stays verbatim.
2. **Merges** — write the merged body to a scratch file, then
`bash "$HOME/.claude/lib/memory.sh" add <kind> --title "<title>"
--body-file <file> --supersedes <first source>` (next-free ID, forward
link, the first source marked superseded); each other source:
`memory.sh status <ID> "superseded by <new-ID>"`. Source bodies stay
verbatim for history. Merged body:
followed by the existing merged-body sub-bullets (Reference lines, caveman pass, frontmatter fields) and the existing item 4 renumbered 3 (inline caveman compression, unchanged). Delete `After each write, regenerate Index from body when rows changed.`
- STEP 4 block: delete the
PREFIX_MAPdeclaration, its comment and the# All body entries have Index rows; no orphansloop (through itsdone), and put in their place:
# Structure: heading shape + unique IDs (v2 registries; a v1 one is named
# and skipped). Same verify the writer runs after every add.
bash "$HOME/.claude/lib/memory.sh" check .claude/memory \
|| echo "CHECK FAILED: fix the lines above before certifying"
-
Report sample
INDEX SANITY: OK→CHECK: OK. -
Failure table row
| Index sanity FAILED at STEP 4 | Print exact missing/orphan IDs. …→| \memory.sh check` FAILED at STEP 4 | Print its culprit lines. Do NOT auto-fix — user re-runs or hand-edits. |`. -
:322-323fixtures listred6-orphan→red6-check. -
Step 4: test-prompts.json id 1
expected→"Skill runs PRECHECK (refuses if working tree dirty on .claude/memory/), audits all 5 registries for A) obsolete/superseded entries B) similar merge candidates C) bloated prose, presents the per-registry PRUNE PLAN with category-level approval gate, applies only approved changes (status lines and merges through lib/memory.sh, never hard-delete), then runs memory.sh check and the fidelity census at STEP 4." -
Step 5: Run, expect green —
make test suite=skills/prune-memory/tests/run-deterministic.sh→ RED-1..7 all GREEN (RED-5 still green: its journal-only sandbox has no registry,checkprints nothing). Run:grep -n 'Index' skills/prune-memory/SKILL.md→ no writer step mentions an Index. -
Step 6: Commit
git add skills/prune-memory
git commit -m "feat(prune-memory): v2 registries — § D dropped, status and merges through memory.sh, STEP 4 verify = memory.sh check"
Task 7: upgrade this repo's registries (the one curation)
Files:
-
Modify:
.claude/memory/{decisions,learnings,blockers,evals}.md(by the lib only) -
Step 1: Dry run on a copy —
d=$(mktemp -d); mkdir -p "$d/.claude/memory"; cp .claude/memory/*.md "$d/.claude/memory/"
bash lib/memory.sh -C "$d" upgrade && bash lib/memory.sh -C "$d" check && echo DRY-OK
Expected: four memory: upgraded … lines, DRY-OK.
- Step 2: Upgrade in place and commit through the chokepoint
bash lib/memory.sh upgrade
bash lib/memory.sh check && echo CHECK-OK
bash lib/memory.sh index blocker | grep -E 'BLK-03[234]'
bash "$HOME/.claude/lib/memory-commit.sh" commit "chore(memory): registries v2 — Index removed, Index-only dates and statuses materialized (memory.sh upgrade)"
Expected: CHECK-OK; BLK-032..034 rows read resolved with a date; memory-commit prints a hash (rc 0: the Index lines sit before the first entry, every other change is an insertion).
- Step 3: Verify the live facts
Run:
make test suite=lib/tests/memory.test.sh→ T12 (live check) green. Run:bash -c '. lib/reconcile.sh; reconcile_blk_open .claude/memory/blockers.md | cut -f1 | tr "\n" " "'→BLK-001 BLK-003 BLK-017 BLK-018 BLK-021 BLK-022 BLK-029(no 032..034). Run:grep -c '^## Index' .claude/memory/*.md→ 0 everywhere.
Final verification (main loop, after Task 7)
make test all suites green; shellcheck lib/memory.sh lib/reconcile.sh lib/memory-commit.sh hooks/session-start.sh; bash lib/memory.sh check; git status --short clean; no file outside this repo changed (git -C <proj> status --short .claude/memory empty for the projects listed in the spec § 2.4 note).