Replace the pm shell-out with the in-process SDK context engine and add --explain - #37
Conversation
pm-context read its items by spawning `pm list-all --json` and parsing the result. That paid a process spawn per invocation and, worse, meant the package hand-rolled relevance scoring that pm-cli already ships and tests. pm-cli 2026.7.26 exposes a real context engine at @unbrained/pm-cli/sdk/query, which had zero adopters across the fleet. This change: - swaps the spawnSync `pm list-all` shell-out for in-process `list()` from @unbrained/pm-cli/sdk/core. Verified equivalent before switching: 24/24 item ids match and the item JSON is equal to the previous shell-out output. - moves relevance and packing onto the SDK's own model rather than a local approximation (scoreContextCandidates / packContextCandidates). - wires context-usage.ts to the SDK context-usage store (recordContextUsageServing / readContextUsageAffinity) instead of hand-parsing runtime/context-usage.jsonl, which pm owns. - adds `pm context-pack --explain`, backed by evaluateContextRanking, so an agent can see WHY each item was selected instead of trusting an opaque pack. Output names the model and every contributing signal. One type fix on top of the above: the local PmItem.parent was `string | undefined` while the SDK's ListResultItem.parent is `string | null | undefined`. Widened the local declaration to match what the SDK actually returns rather than casting the difference away. Verified: typecheck, build, 77/77 tests, changelog:check. Functional proof in a throwaway workspace — the built extension installed into a real pm project and `pm context-pack --status in_progress --explain` returned real per-signal scores (model default-weighted-v1, 9 signals). Refs: pm-context-lk82 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
|
Warning Review limit reached
Next review available in: 48 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Plus Run ID: ⛔ Files ignored due to path filters (8)
📒 Files selected for processing (11)
Walkthrough
ChangesSDK Context Pipeline
Estimated code review effort: 4 (Complex) | ~45 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
@coderabbitai full review @sourcery-ai review Focus areas for this round:
Note: greptile is unavailable for this round — |
🧩 Analysis chain🏁 Script executed: #!/bin/bash
set -euo pipefail
echo '=== PR diff stat ==='
git diff --stat HEAD^ HEAD
echo '=== Changed files ==='
git diff --name-only HEAD^ HEAD
echo '=== Relevant symbol references ==='
rg -n -C 4 \
-e '\breadPmItems\s*\(' \
-e '\bPmItem\b' \
-e '\.parent\b' \
-e '\bscoreContextCandidates\s*\(' \
-e '\bpackContextCandidates\s*\(' \
-e '\bevaluateContextRanking\s*\(' \
-e '\brecordContextUsageServing\s*\(' \
-e '\breadContextUsageAffinity\s*\(' \
-e 'context-usage\.jsonl' \
-e '\bspawnSync\s*\(' \
.
echo '=== Outlines for changed TS/JS source files ==='
git diff --name-only HEAD^ HEAD | while IFS= read -r f; do
case "$f" in
*.ts|*.tsx|*.js|*.mjs|*.cjs)
echo "--- $f ---"
ast-grep outline "$f" --items all || true
;;
esac
done
echo '=== SDK package files / declarations (if present) ==='
fd -HI -t f . node_modules 2>/dev/null | rg '/@unbrained/pm-cli/.*\.(d\.ts|ts|js|mjs|cjs|json)$' | head -n 200 || trueLength of output: 55749 🏁 Script executed: #!/bin/bash
set -euo pipefail
echo '=== Package manifests and locked pm-cli version ==='
fd -HI -t f 'package.json' -o -t f 'package-lock.json' -o -t f 'pnpm-lock.yaml' -o -t f 'yarn.lock' -o -t f 'bun.lockb' . \
| while IFS= read -r f; do
echo "--- $f ---"
rg -n -C 3 '`@unbrained/pm-cli`' "$f" || true
done
echo '=== Source diff: index.ts and context-usage.ts ==='
git diff --unified=80 HEAD^ HEAD -- index.ts context-usage.ts
echo '=== Exact command paths / read / packing / explain / recording ==='
sed -n '1,90p' index.ts
sed -n '820,1085p' index.ts
sed -n '1100,1295p' index.ts
sed -n '1335,1435p' index.ts
echo '=== All direct ledger write APIs and append/write calls ==='
rg -n -C 5 \
-e '\bappendFile(?:Sync)?\s*\(' \
-e '\bwriteFile(?:Sync)?\s*\(' \
-e '\brecordContextUsageServing\s*\(' \
-e 'context-usage\.jsonl' \
--glob '!dist/**' \
.
echo '=== Existing tests and test references for the changed behavior ==='
fd -HI -t f . | rg '(^|/)(test|tests|__tests__|spec)(/|$)|\.(test|spec)\.[cm]?[jt]sx?$' | head -n 250
rg -n -C 3 \
-e '\breadPmItems\s*\(' \
-e '\bcreateSdkPacker\s*\(' \
-e '\bbuildContextExplain\s*\(' \
-e '\bcontext-pack\b' \
-e '\bcontext-usage\b' \
--glob '!dist/**' \
.Length of output: 1724 🏁 Script executed: #!/bin/bash
set -euo pipefail
echo '=== Package manifests and locked pm-cli version ==='
{
fd -HI -t f '^package\.json$' . || true
fd -HI -t f '^package-lock\.json$' . || true
fd -HI -t f '^pnpm-lock\.yaml$' . || true
fd -HI -t f '^yarn\.lock$' . || true
fd -HI -t f '^bun\.lockb$' . || true
} | sort -u | while IFS= read -r f; do
echo "--- $f ---"
rg -n -C 3 '`@unbrained/pm-cli`' "$f" || true
done
echo '=== Source diff: index.ts and context-usage.ts ==='
git diff --unified=80 HEAD^ HEAD -- index.ts context-usage.ts
echo '=== Exact command paths / read / packing / explain / recording ==='
sed -n '1,90p' index.ts
sed -n '820,1085p' index.ts
sed -n '1100,1295p' index.ts
sed -n '1335,1435p' index.ts
echo '=== All direct ledger write APIs and append/write calls ==='
rg -n -C 5 \
-e '\bappendFile(?:Sync)?\s*\(' \
-e '\bwriteFile(?:Sync)?\s*\(' \
-e '\brecordContextUsageServing\s*\(' \
-e 'context-usage\.jsonl' \
--glob '!dist/**' \
.
echo '=== Existing tests and test references for the changed behavior ==='
fd -HI -t f . | rg '(^|/)(test|tests|__tests__|spec)(/|$)|\.(test|spec)\.[cm]?[jt]sx?$' | head -n 250 || true
rg -n -C 3 \
-e '\breadPmItems\s*\(' \
-e '\bcreateSdkPacker\s*\(' \
-e '\bbuildContextExplain\s*\(' \
-e '\bcontext-pack\b' \
-e '\bcontext-usage\b' \
--glob '!dist/**' \
.Length of output: 133750 🏁 Script executed: #!/bin/bash
set -euo pipefail
repo='unbraind/pm-cli'
tag='2026.7.26'
echo '=== Resolve release tag / commit ==='
gh api "repos/${repo}/git/ref/tags/${tag}" --jq '.object.type + " " + .object.sha' \
|| gh api "repos/${repo}/commits/main" --jq '.sha'
echo '=== Locate SDK and CLI source paths ==='
gh api "repos/${repo}/git/trees/main?recursive=1" --paginate --jq \
'.tree[] | select(.type == "blob") | .path' \
| rg -i '(^|/)(sdk|core|query|list|context).*\.(ts|tsx|js|mjs|d\.ts)$|(^|/)package\.json$' \
| head -n 400
echo '=== npm package metadata and exported files ==='
npm view "`@unbrained/pm-cli`@${tag}" version dist.tarball files --json || true
echo '=== Download and inspect the published package without executing repository code ==='
tmp="$(mktemp -d)"
trap 'rm -rf "$tmp"' EXIT
tarball="$(npm view "`@unbrained/pm-cli`@${tag}" dist.tarball)"
curl --fail --silent --show-error --location "$tarball" --output "$tmp/pm-cli.tgz"
tar -tzf "$tmp/pm-cli.tgz" \
| rg -i '(sdk/(core|query)|list|context).*\.(js|mjs|cjs|d\.ts)$|package/package\.json$' \
| head -n 400
tar -xzf "$tmp/pm-cli.tgz" -C "$tmp"
rg -n -C 8 \
-e 'function list\b' \
-e 'const list\b' \
-e 'excludeTerminal' \
-e 'noTruncate' \
-e 'packContextCandidates' \
-e 'recordContextUsageServing' \
-e 'type ContextPackingCandidate' \
-e 'interface ContextPackingCandidate' \
-e 'type ListOptions' \
-e 'interface ListOptions' \
"$tmp/package" \
| head -n 1200Length of output: 201998 🏁 Script executed: #!/bin/bash
set -euo pipefail
tag='2026.7.26'
tmp="$(mktemp -d)"
trap 'rm -rf "$tmp"' EXIT
tarball="$(npm view "`@unbrained/pm-cli`@${tag}" dist.tarball)"
curl --fail --silent --show-error --location "$tarball" --output "$tmp/pm-cli.tgz"
tar -xzf "$tmp/pm-cli.tgz" -C "$tmp"
echo '=== Published SDK entry points ==='
find "$tmp/package" -type f \( -name '*.d.ts' -o -name '*.js' -o -name '*.mjs' \) \
| grep -E '/(sdk|query)(/|[-.])' \
| sed "s#^$tmp/package/##" \
| head -n 250
echo '=== Relevant SDK contracts (declaration files only) ==='
find "$tmp/package" -type f -name '*.d.ts' -print0 \
| xargs -0 rg -n -C 6 \
-e 'interface ContextUsageServingRow' \
-e 'type ContextUsageServingRow' \
-e 'recordContextUsageServing' \
-e 'interface ContextPackingCandidate' \
-e 'type ContextPackingCandidate' \
-e 'packContextCandidates' \
-e 'interface ListResult' \
-e 'type ListResult' \
-e 'excludeTerminal' \
-e 'noTruncate' \
| head -n 800
echo '=== Relevant SDK implementation snippets (bounded per match) ==='
find "$tmp/package" -type f \( -name '*.js' -o -name '*.mjs' \) -print0 \
| xargs -0 rg -l \
-e 'recordContextUsageServing' \
-e 'packContextCandidates' \
-e 'excludeTerminal' \
-e 'noTruncate' \
| while IFS= read -r f; do
echo "--- ${f#$tmp/package/} ---"
rg -n -m 8 -C 3 \
-e 'recordContextUsageServing' \
-e 'packContextCandidates' \
-e 'excludeTerminal' \
-e 'noTruncate' \
"$f" || true
done | head -n 1000
echo '=== Local test coverage around new SDK behavior ==='
rg -n -C 5 \
-e 'readPmItems' \
-e 'noTruncate' \
-e 'excludeTerminal' \
-e 'createSdkPacker' \
-e 'buildContextExplain' \
-e 'included' \
-e 'recordContextUsageServing' \
test index.ts
echo '=== Final line-number anchors ==='
nl -ba index.ts | sed -n '835,855p;990,1028p;1130,1182p;1238,1282p;1355,1385p'
nl -ba context-usage.ts | sed -n '74,100p;268,350p'Length of output: 10218
No duplicate hand-written ledger write remains: the production extension has a single ✅ Action performedFull review finished. Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 31 minutes. |
Run --explain through the same relevance, neighborhood, token-packing, and max-item path as normal context packs. Record every emitted neighbor as included in context-usage while keeping explain non-mutating and preserving legacy list truncation semantics. Add a real SDK-host integration test over a throwaway pm tracker, document the in-process query and serving contract, enforce the max-items ceiling after SDK token optimization, and replace literal NUL source bytes with portable escapes. Repair the previously unreadable pm-context-4apx TOON item with an audited history re-anchor, track the upstream pm-cli round-trip defect as unbraind/pm-cli#747, and record the honest 100/100/100/100 coverage gap as pm-context-3s5f.
|
@coderabbitai full review /gemini review @sourcery-ai review Post-fix review requested for commit c284909. All four prior CodeRabbit findings are addressed, with a real SDK-host integration test covering exact packed/explained membership, max-items enforcement, explain non-mutation, and included neighbor serving rows. Please also inspect the audited TOON tracker recovery and the max-items post-token-budget ceiling. Greptile remains unavailable on the free review limit. |
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
|
✅ Action performedFull review finished. Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 24 minutes. |
|
Acknowledging the Gemini sunset notice at #37 (comment). |
|
Acknowledging the earlier CodeRabbit rate-limit response at #37 (comment). A later full review did complete and produced four actionable findings; all four are fixed in c284909, and a fresh full review has been requested. |
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
|
Resolved every finding in the completed CodeRabbit review at #37 (comment): explain now uses exact final pack membership; every emitted neighbor is recorded included; |
Resolve both post-push DeepScan findings without changing runtime behavior, and record the remediation in pm-context-lk82.
|
Acknowledging the first post-fix Gemini sunset response at #37 (comment). It contains no code finding; the service is unavailable. |
|
Acknowledging CodeRabbit's post-fix response at #37 (comment). It returned no new finding but was rate-limited. Since c66c83f follows, a new full review is requested below as required. |
|
Acknowledging the second Gemini sunset response at #37 (comment). It contains no code finding; this duplicate service notice is recorded and reacted to. |
|
@coderabbitai full review /gemini review @sourcery-ai review Final post-DeepScan review requested for c66c83f. This follow-up only removes the two unused imports reported by DeepScan and records the remediation in the linked pm item. Full release:check remains green at 78/78 tests and zero production audit findings. |
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
|
✅ Action performedFull review finished. Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 19 minutes. |
|
Acknowledging the final-round Gemini sunset response at #37 (comment). No review occurred because the consumer service is retired. |
|
Acknowledging CodeRabbit's final-round response at #37 (comment). It returned no finding but reports review capacity in 19 minutes. All executable checks are green; I will wait within the requested 20-minute window and request one final substantive review when capacity opens. |
|
@coderabbitai full review The stated cooldown has elapsed. Please perform the final substantive review of c66c83f now. All prior findings and both DeepScan findings are resolved; CI, Semgrep, DeepScan, and Hound are green. |
|
✅ Action performedFull review finished. Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 2 seconds. |
|
Acknowledging CodeRabbit final-review initiation at #37 (comment). The check is now genuinely pending; I will wait for the substantive result and address any finding before merge. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 7
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.agents/pm/features/pm-context-lk82.toon:
- Line 3: The feature record still documents superseded SDK APIs and the removed
noTruncate parity behavior. Update its description and line 16 to match the
shipped implementation, using rankContextItems/createSdkPacker,
buildContextExplain/scoreContextItems, and the legacy truncation behavior;
alternatively, clearly mark the earlier API names and noTruncate note as
superseded intermediate details.
In `@index.ts`:
- Around line 1045-1061: Clarify the explain output contract around
buildContextExplain: state that ranks and scores are computed from the supplied
packed subset, not the full corpus or workspace. Add this clarification to the
generated report metadata or the --explain flag description, ensuring readers do
not interpret rank 1 as the workspace-wide top result.
- Around line 1228-1231: Guard the full-corpus ranking in the flow around
rankContextItems and createSdkPacker so rankContextItems runs only when maxItems
is set. Initialize packer as undefined for the no-limit path, and preserve
passing rankedAll to createSdkPacker when maxItems is enabled. Apply the same
conditional change to the corresponding context-handoff logic near the second
occurrence.
- Around line 938-952: Update scoreContextItems to build a single byId Map from
item IDs to items at its start, then use that index when resolving candidates
instead of repeatedly calling byIdOrFail with items.find. Preserve the existing
descriptive CommandError behavior for missing IDs, either through the indexed
lookup helper or equivalent validation.
- Around line 1109-1132: Update resolveSdkRankOptions to handle readSettings or
resolveRuntimeStatusRegistry failures by falling back to the built-in runtime
status registry instead of throwing CommandError, preserving ranking for
malformed settings. Remove the unused _items parameter and update both call
sites to invoke resolveSdkRankOptions without that argument.
In `@README.md`:
- Around line 30-34: Refresh the README “Max items” section to describe
createSdkPacker’s token-budgeted selection, including projection degradation and
--max-items as an additional item-count ceiling. Remove the outdated explanation
that focus items take priority and neighbors are trimmed solely to fit the
remaining count budget.
In `@test/context-pack.test.ts`:
- Around line 602-607: In the context-pack runner invocation, add a brief
comment explaining that omitting author is intentional because it preserves the
ledger state for the final serving assertions. Keep the command options and
expected ledger behavior unchanged.
🪄 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: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: ce81a862-9ff9-4898-8b11-683605fbad0a
⛔ Files ignored due to path filters (8)
dist/context-usage.d.tsis excluded by!**/dist/**dist/context-usage.d.ts.mapis excluded by!**/dist/**,!**/*.mapdist/context-usage.jsis excluded by!**/dist/**dist/context-usage.js.mapis excluded by!**/dist/**,!**/*.mapdist/index.d.tsis excluded by!**/dist/**dist/index.d.ts.mapis excluded by!**/dist/**,!**/*.mapdist/index.jsis excluded by!**/dist/**dist/index.js.mapis excluded by!**/dist/**,!**/*.map
📒 Files selected for processing (11)
.agents/pm/chores/pm-context-4apx.toon.agents/pm/features/pm-context-lk82.toon.agents/pm/history/pm-context-3s5f.jsonl.agents/pm/history/pm-context-4apx.jsonl.agents/pm/history/pm-context-lk82.jsonl.agents/pm/issues/pm-context-3s5f.toonCHANGELOG.mdREADME.mdcontext-usage.tsindex.tstest/context-pack.test.ts
Index relevance candidates once, avoid unnecessary full-corpus ranking, and make emitted-pack-relative explain scores explicit in the API, CLI, docs, and real tracker integration test. Refresh the honest coverage baseline and preserve the remaining 100% gate as tracked work.
|
@coderabbitai full review Please re-review commit 8e5239c and confirm the seven final findings are fully resolved. |
|
/gemini review Please re-review the latest commit 8e5239c. |
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
|
@sourcery-ai review Please re-review the latest commit 8e5239c. |
|
@greptileai please review the latest commit 8e5239c. If this repository is not enabled, please report that explicitly. |
|
Acknowledged the Gemini Code Assist sunset response. No Gemini findings are available for commit 8e5239c; the remaining enabled review and required CI gates continue independently. |
🧩 Analysis chain🏁 Script executed: #!/bin/bash
set -euo pipefail
echo '=== PR metadata and recent commits ==='
gh pr view 37 --repo unbraind/pm-context --json number,headRefOid,baseRefOid,commits \
--jq '{number, headRefOid, baseRefOid, commits: [.commits[] | {oid, messageHeadline}]}'
echo '=== Commit 8e5239c metadata and changed files ==='
git show --no-ext-diff --format=fuller --stat 8e5239c
git diff-tree --no-commit-id --name-status -r 8e5239c
echo '=== Commit diff ==='
git show --no-ext-diff --format= --unified=80 8e5239c
echo '=== Recent PR review/comment bodies (to identify the seven final findings) ==='
gh api --paginate 'repos/unbraind/pm-context/issues/37/comments?per_page=100' \
--jq '.[] | select(.user.login == "coderabbitai[bot]" or .user.login == "coderabbitai") | {created_at, body}' \
| tail -n 80
gh api --paginate 'repos/unbraind/pm-context/pulls/37/comments?per_page=100' \
--jq '.[] | select(.user.login == "coderabbitai[bot]" or .user.login == "coderabbitai") | {created_at, path, line, body}' \
| tail -n 100
echo '=== Outline and focused symbols in changed TypeScript files ==='
git diff-tree --no-commit-id --name-only -r 8e5239c \
| while IFS= read -r f; do
case "$f" in
*.ts|*.tsx|*.js|*.mjs|*.cjs)
echo "--- outline: $f ---"
ast-grep outline "$f" --items all || true
;;
esac
done
echo '=== Focused implementation and test references ==='
rg -n -C 5 \
-e '\breadPmItems\s*\(' \
-e '\bcreateSdkPacker\s*\(' \
-e '\bbuildContextExplain\s*\(' \
-e '\brecordPackServing\s*\(' \
-e '\brecordContextUsageServing\s*\(' \
-e '\bmaxItems\b' \
-e '\bincluded\s*:' \
-e '\bnoTruncate\b' \
-e '\bexcludeTerminal\b' \
index.ts context-usage.ts test 2>/dev/null || trueLength of output: 202078 🏁 Script executed: #!/bin/bash
set -euo pipefail
echo '=== Source-only files changed by 8e5239c ==='
git diff-tree --no-commit-id --name-only -r 8e5239c \
| rg -v '^(dist/|.*\.map$)' \
| sort
echo '=== Exact feature-record text ==='
sed -n '1,120p' .agents/pm/features/pm-context-lk82.toon
echo '=== README context-engine and max-items sections ==='
rg -n -C 6 -e 'SDK' -e 'shell' -e '[Mm]ax items' -e 'projection' README.md
echo '=== Final implementation regions ==='
nl -ba index.ts | sed -n '827,855p;920,963p;985,1108p;1110,1170p;1218,1270p;1346,1376p'
nl -ba test/context-pack.test.ts | sed -n '558,652p'
echo '=== Obsolete implementation/documentation terms in non-generated files ==='
rg -n -C 2 \
-e '\bscoreContextCandidates\b' \
-e '\bpackContextCandidates\b' \
-e '\bevaluateContextRanking\b' \
-e '\bnoTruncate\b' \
-e 'spawnSync\("pm"' \
--glob '!dist/**' \
--glob '!*.map' \
.
echo '=== Diff limited to the seven reviewed source/doc files ==='
git diff --no-ext-diff --check 8e5239c^ 8e5239c -- \
.agents/pm/features/pm-context-lk82.toon README.md index.ts test/context-pack.test.ts
git diff --no-ext-diff --unified=12 8e5239c^ 8e5239c -- \
.agents/pm/features/pm-context-lk82.toon README.md index.ts test/context-pack.test.tsLength of output: 9842 🏁 Script executed: #!/bin/bash
set -euo pipefail
echo '=== Final source: list equivalence, scoring index, packing, explain ==='
sed -n '827,1108p' index.ts
echo '=== Final source: rank options, serving, context-pack and handoff command paths ==='
sed -n '1110,1170p' index.ts
sed -n '1218,1270p' index.ts
sed -n '1346,1376p' index.ts
echo '=== Final integration test ==='
sed -n '558,652p' test/context-pack.test.ts
echo '=== Focused source search with line anchors ==='
rg -n -C 3 \
-e 'const byId = new Map' \
-e 'function byIdOrFail' \
-e 'ranking_scope' \
-e 'Ranking scope:' \
-e 'resolveSdkRankOptions\(' \
-e 'if \(maxItems\)' \
-e 'Intentionally omit author' \
index.ts test/context-pack.test.ts
echo '=== Non-generated stale-term search ==='
rg -n -C 2 \
-e '\bscoreContextCandidates\b' \
-e '\bevaluateContextRanking\b' \
-e '\bnoTruncate\b' \
-e 'read-only and shells out' \
--glob '!dist/**' \
--glob '!*.map' \
. || true
echo '=== Commit hygiene ==='
git diff --no-ext-diff --check 8e5239c^ 8e5239c
git diff --no-ext-diff --unified=12 8e5239c^ 8e5239c -- \
.agents/pm/features/pm-context-lk82.toon README.md index.ts test/context-pack.test.tsLength of output: 55890
I also confirmed the SDK list call preserves default truncation semantics, the feature history labels its obsolete ✅ Action performedFull review finished. Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 48 minutes. |
|
Acknowledged. Thank you for confirming all seven dispositions on 8e5239c, including the malformed-settings fallback rationale, and reporting no additional actionable findings. The full local 78/78 release gate and all enabled GitHub checks are green; the explicit coverage gap remains tracked in pm-context-3s5f. |
1 participant
What this changes
pm-contextnow reads, ranks, packs, explains, and records workspace context through pm-cli's typed in-process SDK rather than parsing apm list-all --jsonsubprocess. The implementation uses the installed 2026.7.26 SDK as the single query/relevance/context-usage authority.SDK behavior
list({ full: true, includeBody: true, excludeTerminal: false }, { pmRoot }), preserving the oldlist-all --include-bodytruncation semantics.--max-itemscount ceiling after SDK token optimization.--explainthrough the exact normal filter, relevance, neighborhood, token-budget, and count-ceiling path, then explains only the emitted membership.included: true;--explainremains non-mutating.Review hardening
All four CodeRabbit findings from the full review were reproduced and fixed. A real SDK-host integration test initializes a throwaway pm tracker, creates dependency neighbors through the typed SDK, and proves:
--max-items;--max-items 1emits exactly one item;README wording now documents the in-process dependency and intentional best-effort serving side effect. Literal NUL bytes in TypeScript source were replaced by behavior-identical
\u0000escapes so repository tooling treats every source file as text.PM storage recovery
pm validatefound a previously committed item that pm 2026.7.26 could no longer decode because the canonical TOON encoder emits a quotednotes[] ... key:valuescalar its decoder rejects. The smallest text-only recovery was followed by an auditedpm history-repair; verification and storage integrity are green again. Core round-trip and recovery-path evidence is upstream in pm-cli#747.Verification
The required 100/100/100/100 source-coverage gate is explicitly tracked rather than falsely claimed.
PM items
Additional SDK issues discovered during ecosystem adoption: pm-cli#738, pm-cli#740, and pm-cli#747.