@llipe.com/dev-tasks 0.17.0 → 0.17.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -22,11 +22,41 @@
22
22
  #
23
23
  # Rules 1 and 4 read each command's arguments from its own simple-command
24
24
  # segment (split at `&&`, `||`, `;`, `|`, `&`, and newlines outside quotes —
25
- # see command_segments()), never from the rest of the command string. A
25
+ # see executed_view()), never from the rest of the command string. A
26
26
  # chained `gh pr create --base main` is not a push to `main`, a `--tags` in a
27
27
  # commit message is not a tag push, and every push or merge in a chain is
28
28
  # checked, not only the last one (issue #256).
29
29
  #
30
+ # Executed view (issue #339). Rules 1 and 4 judge only text that bash runs:
31
+ # - Quoted heredocs (`<<'ID'`, `<<"ID"`, `<<-'ID'`, ID a plain identifier)
32
+ # are read as heredocs. Their body lines are never parsed for quotes or
33
+ # separators, so an apostrophe in a body cannot glue later commands
34
+ # together. An unquoted delimiter (`<<EOF`) and a `<<<` here-string are
35
+ # expanded by bash and are never relaxed.
36
+ # - A command is relaxed only when every segment's first word is on the
37
+ # allow-list `git`, `gh`, `echo`, `printf`, `cat`. Then heredoc bodies and
38
+ # `echo`/`printf`/`cat` arguments are ignored; `git` and `gh` arguments
39
+ # are always kept.
40
+ # - Anything that could run code keeps the whole command strict (the
41
+ # pre-#339 reading, where heredoc body lines are read as commands, so
42
+ # `bash <<'EOF'` stays blocked): `$(`, a backtick, `$[`, `${`, `<(`, `>(`,
43
+ # an assignment prefix (`X=1 cmd`), or a runner or interpreter word
44
+ # anywhere (`bash`, `sh`, `eval`, `source`, `.`, `env`, `command`, `exec`,
45
+ # `xargs`, `sudo`, `nohup`, `time`, `python*`, `node`, `fish`, ...).
46
+ # Triggers inside a quoted heredoc body are ignored, since bash never
47
+ # expands it; single-quoted text elsewhere still counts.
48
+ # - Also strict, failing toward the old reading on any doubt: `$'`, `!`,
49
+ # a comment (`#` at a word start), any other `<<`, an open quote, an
50
+ # unterminated heredoc, two heredocs on one line, `printf -v`, `git -c`,
51
+ # `--config-env` or `--exec-path`, the plain text `.git/`, `hook` or
52
+ # `.husky` (a text match, not a path check: it narrows writing a hook
53
+ # that runs later, which only branch protection closes), and text fed by
54
+ # a pipe or heredoc to anything but `git commit`,
55
+ # `gh issue` or `gh pr` (an alias or extension cannot shadow a built-in).
56
+ # Rules 2 (Conventional Commits) and 3 (inline `--body`) deliberately read
57
+ # argument text and do not use the executed view. `--body-file <path>`
58
+ # remains the recommended way to pass an issue or PR body.
59
+ #
30
60
  # 5. (Issue #178) The same "no write to the default branch" invariant as
31
61
  # rule 1, enforced on the mutating GitHub MCP tool surface, not just
32
62
  # `Bash`. Before this rule, `.claude/settings.json` matched `"Bash"`
@@ -222,51 +252,194 @@ tokenize() {
222
252
  printf '%s' "$1" | xargs -n1 -- printf '%s\n' 2>/dev/null
223
253
  }
224
254
 
225
- # Split the raw command into simple-command segments, one per output line,
226
- # at `&&`, `||`, `;`, `|`, `&`, and newlines outside single or double quotes
227
- # (backslash escapes honored). Every rule that reads a command's arguments
228
- # reads them from its own segment, never from the rest of the string. Before
229
- # this, `git push -u origin feat && gh pr create --base main` read `main` as
230
- # a push destination, `--tags` anywhere in the string read as a tag push, and
231
- # a greedy `.*git push` match checked only the LAST push in a chained
232
- # command, so `git push origin main && git push origin feat` was allowed
233
- # (issue #256). Portable awk only: macOS ships BSD awk, not gawk.
234
- command_segments() {
255
+ # Build the executed view of the raw command (issue #339). Prints the mode,
256
+ # `relaxed` or `strict`, on line one, then one simple-command segment per
257
+ # line. Segments split at `&&`, `||`, `;`, `|`, `&`, and newlines outside
258
+ # single or double quotes (backslash escapes honored; `>&`, `<&` and `&>` are
259
+ # redirections, not separators), so every rule that reads a command's
260
+ # arguments reads them from its own segment, never from the rest of the
261
+ # string (issue #256).
262
+ #
263
+ # Quoted heredocs (`<<'ID'`, `<<"ID"`, `<<-'ID'`, ID a plain identifier) are
264
+ # read as heredocs: their body lines are never parsed for quotes or
265
+ # separators, so an apostrophe in a body (`don't`) cannot open a phantom quote
266
+ # that glues later commands together.
267
+ #
268
+ # relaxed: every segment's first word is `git`, `gh`, `echo`, `printf` or
269
+ # `cat`, and nothing in the command can run code (see the header for the
270
+ # triggers). Heredoc bodies are dropped and `echo`/`printf`/`cat` segments
271
+ # are cut to their first word, since bash never runs that text. `git` and
272
+ # `gh` segments are kept whole.
273
+ # strict: anything else, or any doubt. Every segment is kept, and each heredoc
274
+ # body line is a segment of its own (backslash-newline joined), so
275
+ # `bash <<'EOF'` with a push in the body stays blocked. This is the
276
+ # pre-#339 view except for the heredoc body handling.
277
+ #
278
+ # Portable awk only: macOS ships BSD awk, not gawk.
279
+ executed_view() {
235
280
  printf '%s\n' "$cmd" | awk '
281
+ function flush() {
282
+ k++; seg[k] = cur; body[k] = 0; fed[k] = cur_fed
283
+ cur = ""; cur_fed = 0; started = 0
284
+ }
285
+ function emit(ch) {
286
+ if (!started && ch != " " && ch != "\t") {
287
+ started = 1; cur_fed = pipe_pending; pipe_pending = 0
288
+ }
289
+ cur = cur ch
290
+ }
291
+ function add_body(line) { k++; seg[k] = line; body[k] = 1 }
292
+ # Read the pending heredoc bodies that start after the newline at i.
293
+ function read_bodies( p, m, e, line, cmp, closed, acc) {
294
+ p = i + 1
295
+ for (m = 1; m <= npend; m++) {
296
+ closed = 0; acc = ""
297
+ while (p <= n) {
298
+ e = index(substr(s, p), "\n")
299
+ if (e == 0) { line = substr(s, p); p = n + 1 }
300
+ else { line = substr(s, p, e - 1); p = p + e }
301
+ cmp = line
302
+ if (pdash[m]) sub(/^\t+/, "", cmp)
303
+ if (cmp == pid[m]) { closed = 1; break }
304
+ if (line ~ /\\$/) { acc = acc substr(line, 1, length(line) - 1); continue }
305
+ add_body(acc line); acc = ""
306
+ }
307
+ if (acc != "") add_body(acc)
308
+ if (!closed) { strict = 1; break }
309
+ }
310
+ npend = 0
311
+ i = p - 1
312
+ }
236
313
  { s = s (NR > 1 ? "\n" : "") $0 }
237
314
  END {
238
- out = ""; q = ""; n = length(s)
315
+ n = length(s); q = ""; prev = ""; k = 0; cur = ""; started = 0
316
+ cur_fed = 0; pipe_pending = 0; npend = 0; line_openers = 0; strict = 0
239
317
  for (i = 1; i <= n; i++) {
240
- c = substr(s, i, 1)
318
+ c = substr(s, i, 1); c2 = substr(s, i, 2)
319
+ # Strictness triggers, anywhere outside a quoted heredoc body
320
+ # (bodies are consumed by read_bodies and never reach this loop).
321
+ if (c2 == "$(" || c2 == "$[" || c2 == "${" || c2 == "$\047" ||
322
+ c2 == "<(" || c2 == ">(" || c2 == "<<" || c == "`" || c == "!") {
323
+ if (!(q == "" && c2 == "<<")) strict = 1
324
+ }
241
325
  if (q != "") {
242
326
  # Inside quotes a newline is part of the argument, never a
243
327
  # separator: map it to a space so the segment stays whole.
244
- if (c == "\n") { out = out " "; continue }
245
- if (c == q) { q = ""; out = out c; continue }
328
+ if (c == "\n") { if (npend) strict = 1; emit(" "); continue }
329
+ if (c == q) { q = ""; emit(c); prev = c; continue }
246
330
  if (c == "\\" && q == "\"" && i < n) {
247
331
  d = substr(s, i + 1, 1); i++
248
- if (d != "\n") out = out c d # backslash-newline is a continuation
332
+ if (d != "\n") { emit(c); emit(d) } # backslash-newline is a continuation
249
333
  continue
250
334
  }
251
- out = out c
335
+ emit(c)
252
336
  continue
253
337
  }
254
338
  if (c == "\\" && i < n) {
255
339
  d = substr(s, i + 1, 1); i++
256
- if (d != "\n") out = out c d # backslash-newline is a continuation
340
+ # An escaped character is literal: prev must not read it as the `>`
341
+ # or `<` of a `>&`/`<&` redirection (`\>& cmd` backgrounds cmd).
342
+ if (d != "\n") { emit(c); emit(d); prev = c } # backslash-newline is a continuation
343
+ continue
344
+ }
345
+ # A word-start `#` opens a comment: keep the pre-#339 reading.
346
+ if (c == "#" && (prev == "" || prev ~ /[ \t\n;&|()]/)) strict = 1
347
+ if (c == "\"" || c == "\047") { q = c; emit(c); prev = c; continue }
348
+ if (c2 == "<<") {
349
+ if (substr(s, i + 2, 1) == "<") { # a here-string
350
+ strict = 1; emit("<"); emit("<"); emit("<"); i += 2; prev = "<"; continue
351
+ }
352
+ j = i + 2; dash = 0; ok = 0
353
+ if (substr(s, j, 1) == "-") { dash = 1; j++ }
354
+ while (substr(s, j, 1) == " " || substr(s, j, 1) == "\t") j++
355
+ oq = substr(s, j, 1)
356
+ if (oq == "\047" || oq == "\"") {
357
+ e = index(substr(s, j + 1), oq)
358
+ if (e > 0) {
359
+ id = substr(s, j + 1, e - 1); after = substr(s, j + e + 1, 1)
360
+ if (id ~ /^[A-Za-z_][A-Za-z0-9_]*$/ && (after == "" || after ~ /[ \t\n;&|<>)]/)) ok = 1
361
+ }
362
+ }
363
+ if (!ok) { strict = 1; emit(c); prev = c; continue } # unquoted, arithmetic, ...
364
+ if (++line_openers > 1) strict = 1 # two heredocs on one line
365
+ npend++; pid[npend] = id; pdash[npend] = dash
366
+ for (m = i; m <= j + e; m++) emit(substr(s, m, 1))
367
+ cur_fed = 1 # the heredoc feeds this segment
368
+ i = j + e; prev = oq
257
369
  continue
258
370
  }
259
- if (c == "\"" || c == "\047") { q = c; out = out c; continue }
260
- if (c == ";" || c == "|" || c == "&" || c == "\n") { out = out "\n"; continue }
261
- out = out c
371
+ if (c == "&" && (prev == ">" || prev == "<" || substr(s, i + 1, 1) == ">")) {
372
+ emit(c); prev = c; continue
373
+ }
374
+ if (c == "|" && substr(s, i + 1, 1) == "|") { # `||` is not a pipe
375
+ flush(); i++; prev = c; continue
376
+ }
377
+ if (c == ";" || c == "|" || c == "&" || c == "\n") {
378
+ flush(); prev = c
379
+ if (c == "|") pipe_pending = 1
380
+ if (c == "\n") { line_openers = 0; if (npend) read_bodies() }
381
+ continue
382
+ }
383
+ emit(c); prev = c
384
+ }
385
+ if (q != "" || npend) strict = 1 # an open quote or an unread heredoc
386
+ flush()
387
+
388
+ split("git gh echo printf cat", a, " ")
389
+ for (m in a) allow[a[m]] = 1
390
+ split("bash sh zsh ksh dash ash mksh fish csh tcsh busybox eval source . exec env command builtin xargs sudo doas su nohup time nice ionice timeout setsid stdbuf unbuffer watch parallel script chroot ssh trap node nodejs deno bun perl ruby php lua luajit osascript awk gawk mawk nawk sed tclsh wish expect", a, " ")
391
+ for (m in a) runner[a[m]] = 1
392
+ text = ""
393
+ for (m = 1; m <= k; m++) {
394
+ if (body[m]) continue
395
+ t = seg[m]; sub(/^[ \t]+/, "", t)
396
+ if (t == "") continue
397
+ text = text " " t
398
+ nw = split(t, w, /[ \t]+/)
399
+ if (!(w[1] in allow)) strict = 1
400
+ # Text fed to git or gh is data only for these built-ins; an alias
401
+ # or extension cannot shadow a built-in.
402
+ if (fed[m] && w[1] == "git" && w[2] != "commit") strict = 1
403
+ if (fed[m] && w[1] == "gh" && w[2] != "issue" && w[2] != "pr") strict = 1
404
+ if (w[1] == "printf" && w[2] == "-v") strict = 1
405
+ if (w[1] == "git") {
406
+ for (x = 2; x <= nw; x++) if (w[x] == "-c" || w[x] ~ /^--(config-env|exec-path)/) strict = 1
407
+ }
408
+ }
409
+ low = tolower(text)
410
+ if (index(low, ".git/") || index(low, "hook") || index(low, ".husky")) strict = 1
411
+ nt = split(text, tok, /[^A-Za-z0-9_.+-]+/)
412
+ for (x = 1; x <= nt; x++) {
413
+ if ((tok[x] in runner) || tok[x] ~ /^(python|pypy)[0-9.]*$/) strict = 1
414
+ }
415
+
416
+ if (strict) {
417
+ print "strict"
418
+ for (m = 1; m <= k; m++) print seg[m]
419
+ exit
420
+ }
421
+ print "relaxed"
422
+ for (m = 1; m <= k; m++) {
423
+ if (body[m]) continue
424
+ t = seg[m]; sub(/^[ \t]+/, "", t)
425
+ split(t, w, /[ \t]+/)
426
+ if (w[1] == "echo" || w[1] == "printf" || w[1] == "cat") print w[1]
427
+ else print seg[m]
262
428
  }
263
- print out
264
429
  }'
265
430
  }
266
431
 
267
- # Segments matching the extended regex $1, whitespace-normalized, one per line.
432
+ # The executed view, computed once: its segments (the mode line dropped), and
433
+ # the segments joined into one whitespace-normalized line for whole-command
434
+ # checks.
435
+ _exec_out="$(executed_view)"
436
+ exec_segments="$(printf '%s\n' "$_exec_out" | tail -n +2)"
437
+ exec_norm="$(printf '%s' "$exec_segments" | tr '\n' ' ' | tr -s ' ')"
438
+
439
+ # Executed-view segments matching the extended regex $1, whitespace-normalized,
440
+ # one per line.
268
441
  segments_matching() {
269
- command_segments | tr -s ' \t' ' ' | grep -E -- "$1"
442
+ printf '%s\n' "$exec_segments" | tr -s ' \t' ' ' | grep -E -- "$1"
270
443
  }
271
444
 
272
445
  # Argument tokens following EACH `git push` in segment $1, one per line.
@@ -482,7 +655,7 @@ gh_pr_merge_target() {
482
655
 
483
656
  # `git push` targeting the default branch (resolved dynamically; not hardcoded
484
657
  # to `main` so a repo on `master`/`trunk` is protected identically).
485
- if printf '%s' "$norm" | grep -Eq "$_push_segment_re|[;&|]git +push"; then
658
+ if printf '%s' "$exec_norm" | grep -Eq "$_push_segment_re|[;&|]git +push"; then
486
659
  default_branch="$(resolve_default_branch)"
487
660
  if push_targets_ref "$default_branch"; then
488
661
  block "pushing to '$default_branch' is not allowed. Open a PR; only the user may merge into $default_branch."
@@ -490,7 +663,7 @@ if printf '%s' "$norm" | grep -Eq "$_push_segment_re|[;&|]git +push"; then
490
663
  fi
491
664
 
492
665
  # `git merge`: two independent checks.
493
- if printf '%s' "$norm" | grep -Eq "${_cmd_prefix}git +merge([[:space:]]|$)|[;&|]git +merge([[:space:]]|$)"; then
666
+ if printf '%s' "$exec_norm" | grep -Eq "${_cmd_prefix}git +merge([[:space:]]|$)|[;&|]git +merge([[:space:]]|$)"; then
494
667
  current_branch="$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo '')"
495
668
  default_branch="$(resolve_default_branch)"
496
669
 
@@ -637,7 +810,7 @@ if push_is_tag_push; then
637
810
  fi
638
811
 
639
812
  # Block `gh release create` (cutting a release implies creating a tag).
640
- if printf '%s' "$norm" | grep -Eq 'gh +release +create'; then
813
+ if printf '%s' "$exec_norm" | grep -Eq 'gh +release +create'; then
641
814
  block "cutting a GitHub release is human-only (it creates a tag). See github-ops 'Tags'."
642
815
  fi
643
816
 
package/AGENTS.md CHANGED
@@ -97,6 +97,8 @@ Kiro (`.kiro/hooks/`) and Claude Code (`.claude/hooks/`, wired via `.claude/sett
97
97
 
98
98
  Hook enforcement is best-effort and fails open on unexpected errors, with one exception: if `git-guard` cannot verify a `gh pr merge`'s (or its MCP equivalent's) base branch — the `gh pr view` lookup fails (`gh` missing, unauthenticated, or a network error), or an MCP merge/auto-merge call omits enough `tool_input` to identify the PR at all — it fails **closed** and blocks the call, since an unverified base could be the default branch.
99
99
 
100
+ The Claude Code `git-guard` judges pushes, merges and tags only by commands that run: it ignores quoted-heredoc text and `printf`/`echo`/`cat` text when it can show that text is never executed, and `--body-file <path>` remains the recommended way to pass an issue or PR body.
101
+
100
102
  **Hooks are advisory, not the gate.** Every hook in this table is a best-effort, local, pattern-matching check over a command string or a structured tool-input shape — not a real parser, and not a completeness guarantee against every future command form or tool surface (a new MCP server, a future first-party tool) that could reopen the same class of gap. **GitHub branch protection on the default branch — require a PR, require ≥1 approving review, require passing status checks, disable force-push and branch deletion — is the actual, server-side, unbypassable-by-any-tool-surface control.** Hooks narrow the window and catch the common cases; they do not replace branch protection. Configure it per the README "Configure branch protection" step and `activity-init`'s Repository Setup section.
101
103
 
102
104
  ## General Agent Guidelines
@@ -97,6 +97,8 @@ Kiro (`.kiro/hooks/`) and Claude Code (`.claude/hooks/`, wired via `.claude/sett
97
97
 
98
98
  Hook enforcement is best-effort and fails open on unexpected errors, with one exception: if `git-guard` cannot verify a `gh pr merge`'s (or its MCP equivalent's) base branch — the `gh pr view` lookup fails (`gh` missing, unauthenticated, or a network error), or an MCP merge/auto-merge call omits enough `tool_input` to identify the PR at all — it fails **closed** and blocks the call, since an unverified base could be the default branch.
99
99
 
100
+ The Claude Code `git-guard` judges pushes, merges and tags only by commands that run: it ignores quoted-heredoc text and `printf`/`echo`/`cat` text when it can show that text is never executed, and `--body-file <path>` remains the recommended way to pass an issue or PR body.
101
+
100
102
  **Hooks are advisory, not the gate.** Every hook in this table is a best-effort, local, pattern-matching check over a command string or a structured tool-input shape — not a real parser, and not a completeness guarantee against every future command form or tool surface (a new MCP server, a future first-party tool) that could reopen the same class of gap. **GitHub branch protection on the default branch — require a PR, require ≥1 approving review, require passing status checks, disable force-push and branch deletion — is the actual, server-side, unbypassable-by-any-tool-surface control.** Hooks narrow the window and catch the common cases; they do not replace branch protection. Configure it per the README "Configure branch protection" step and `activity-init`'s Repository Setup section.
101
103
 
102
104
  ## General Agent Guidelines
package/CHANGELOG.md CHANGED
@@ -5,7 +5,26 @@ All notable changes to this project will be documented in this file.
5
5
  The format is based on [Keep a Changelog](https://keepachangelog.com/),
6
6
  and this project adheres to [Semantic Versioning](https://semver.org/).
7
7
 
8
- ## [Unreleased]
8
+ ## [0.17.1] - 2026-11-03
9
+
10
+ ### Fixed
11
+
12
+ - fix(hooks): treat an escaped redirect character as literal in git-guard
13
+ - fix(hooks): judge git-guard pushes by executed commands, not heredoc text
14
+ - fix(release): attach the bundle to an existing release; record 0.17.0 (#337)
15
+
16
+ ### Changed
17
+
18
+ - Merge pull request #344 from llipe/chore/pre-release-format
19
+ - chore: apply prettier to docs and archived planner state
20
+ - Merge pull request #341 from llipe/issue/339-git-guard-heredoc-push
21
+ - test(hooks): isolate each executed-view trigger and document the runbook
22
+ - test(hooks): pin git-guard executed-view cases for #339
23
+ - Merge pull request #336 from llipe/chore/projects-tracking-phase-3-closeout
24
+ - Merge pull request #338 from llipe/chore/337-release-bundle-existing-release
25
+ - chore(roadmap): close out projects-tracking-phase-3
26
+
27
+ ## [0.17.0] - 2026-10-27
9
28
 
10
29
  ### Added
11
30
 
package/TESTING.md CHANGED
@@ -1,5 +1,5 @@
1
1
  ---
2
- version: 1.12
2
+ version: 1.13
3
3
  name: Testing Standard
4
4
  description: Canonical testing contract for dev-tasks — declares test layers, runners, commands, fixtures, and coverage policy.
5
5
  status: filled
@@ -154,42 +154,43 @@ The test suite runs to completion, and its only known intermittent failure is th
154
154
 
155
155
  Behavioural acceptance criteria of the projects-tracking PRD cannot run inside Vitest: a mocked `gh` would test the mock, not the agent (D-39). Each is verified by reading GitHub after a real event, using the `DF-*` rows of `workstream/test-plan-projects-tracking-phase-1.md`, and recorded here with the satisfying issue number. `pending` means the event has not happened yet; `n/a (pre-rule)` marks the two wave 1 stories that ran before S-003's status-comment rule existed in the loaded prompt (D-42).
156
156
 
157
- | Criterion | Event | Issue | Status |
158
- | ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
159
- | PRD AC-1 — publish creates the PRD/wave/story hierarchy | Wave 2 publish by `product-engineer` (DF-1, zero-match path) | #289, #290, #291 to #296 | observed 2026-10-21: wave 2 publish created the PRD and wave tracking issues and attached six stories, hierarchy verified(7 of 7) |
160
- | PRD AC-2 (live half) — tracking issues carry `level:` only | Wave 2 publish (DF-1) | #289, #290 | observed 2026-10-21: both tracking issues carry their `level:` label only |
161
- | PRD AC-3 (b) — story status comment, S-001 | `/developer` on #270 (DF-2) | #270 | n/a (pre-rule) |
162
- | PRD AC-3 (b) — story status comment, S-002 | `/developer` on #271 (DF-2) | #271 | n/a (pre-rule) |
163
- | PRD AC-3 (b) — story status comment, S-003 | `/developer` on #272 (DF-2) | #272 | observed 2026-10-20: one marked comment (id 6302983274), edited in place; comment created at sub-task 3.2a, after the draft PR, because the rule did not exist before 3.2 (S-003 DR-2) |
164
- | PRD AC-3 (b) — story status comment, S-004 | `/developer` on #273 (DF-2) | #273 | observed 2026-10-20: one marked comment (id 6303270800), edited in place |
165
- | PRD AC-3 (b) — story status comment, S-005 | `/developer` on #274 (DF-2) | #274 | observed 2026-10-20: one marked comment (id 6303586826), edited in place |
166
- | PRD AC-3 (b) — story status comment, S-006 | `/developer` on #275 (DF-2) | #275 | observed 2026-10-21: one marked comment (id 6305768927), edited in place |
167
- | PRD AC-3 (a, live half) — Project Status written only by automations | Consolidated PR to main opens with `Closes #N` (Status → In progress) and merges (→ Done); story PRs into the integration branch move nothing (D-50) (DF-5) | #270 to #275; #285, #290 to #296 | observed: wave 1 Done after PR #282 merged; wave 2 In progress for all eight issues at 20:39Z 2026-10-22 after consolidated PR #312 opened, read with `projectItems`, no agent write; Done for all eight at 21:00Z after PR #312 merged, `project_observation: observed` |
168
- | FR-24 — wave update, no-parent path (404) | Any wave 1 story's `/developer` run (DF-3) | #272, #273, #274, #275 | observed: `gh api …/parent` returns 404; closeout carries `wave_status: skipped(no parent)` |
169
- | FR-24 — wave update, success path (200) | First main-thread `/developer` run on a wave 2 story: parent found, wave comment edited (DF-3) | pending | pending |
170
- | D-36 — second publish reuses the PRD issue (one-match path) | Wave 3 publish (DF-1b) | #289, #317 | observed 2026-10-24: the wave 3 publish found exactly one open `[PRD-projects-tracking]` issue and reused #289, then created wave #317 under it (`workstream/github-publication-projects-tracking-phase-3.md`). |
171
- | PRD AC-5 (live half) — runbook exercise (DF-4) | Owner follows `docs/runbooks/runbook-configure-github-project.md` on `llipe/dev-tasks` | #274 (Project 5) | observed 2026-10-21: Project 5 created from the Team planning template; seven shipped workflows enabled; auto-add `is:issue`; fields kept by decision D-48 |
172
- | Wave 2 DF-6 — checkpoint committed after every merge | `/planner` wave 2 run (S-008 rule) | integration/projects-tracking-phase-2 | observed 2026-10-22, hand-applied (rule merged at S-008): six `chore(planner): checkpoint after S-0xx merge` commits, each after its merge |
173
- | Wave 2 DF-7 — state file carries `Wave issue` and `PRD issue` | `/planner` wave 2 run (S-008, S-018 rules) | #290, #289 | observed 2026-10-22, hand-applied |
174
- | Wave 2 DF-8 — one wave status comment, edited in place | `/planner` wave 2 run (S-009 rule) | #290 | partly observed 2026-10-22, hand-applied: one marker comment edited at run start and every merge; never edited at a story's PR-ready closeout, so no story showed `PR open` (rollup RD-1) |
175
- | Wave 2 DF-9 — no per-merge table comments | `/planner` wave 2 run (S-009 rule) | #290 to #296 | observed 2026-10-22 |
176
- | Wave 2 DF-11 — consolidated PR carries `Closes` for every merged story and the wave | Consolidated PR open (S-010 rule) | #312 | observed 2026-10-22: `Closes` for #291 to #296, #290 and #285 under `## Why`; GitHub links all eight |
177
- | Wave 2 DF-15 — close-out runs from the merged prompt | Phase 6 in a fresh session on `main` (S-011 rule) | close-out PR | observed 2026-10-22: Phase 6 run from the merged prompt on `main` |
178
- | Wave 2 DF-16 — PRD status comment at run start, PR open and close-out | `/planner` wave 2 run (S-018 rule) | #289 | run start hand-applied; PR open and shipped observed 2026-10-22 from the merged prompt |
179
- | DF-21 — S-016 roadmap check is a no-op on a canonical roadmap | Wave 3 migration, before `--plan` | #323 | observed 2026-10-25: `roadmap-parity` 28/28; this repository's roadmap was already canonical, so no conversion PR. |
180
- | DF-22 — `--plan` is read-only (FR-17) | Wave 3 migration, `--plan` | #323 | observed 2026-10-26: labels and open issues identical before and after a `--plan` run (qa-engineer, PR #332). |
181
- | DF-23 — Migration Mode asks one question per review item (S-017/AC-2) | Wave 3 migration review walk | #323 | observed 2026-10-25 (planner attestation): nine items asked one per message; the owner overrode two recommendations (#108, #262). |
182
- | DF-24 — `Approved: yes` set only on explicit approval (D-80) | Before `--apply` | #323 | observed 2026-10-25 (planner attestation): set after the owner wrote "approved"; one `Approved: yes` line; Plan digest recomputes to the header. |
183
- | DF-25 — `--apply` records every write (S-015/AC-3, AC-8) | Wave 3 migration, `--apply` | #323 | observed 2026-10-25: exit 0, 14 done, 0 skipped, 0 failed; every row matches GitHub. |
184
- | DF-26 — the Project batch is the owner's (PRD AC-12, D-81) | Owner runs batch 7 | #323 | observed 2026-10-26: four steps ticked by the owner (screenshot: auto-add `is:issue`, Item closed and Pull request merged on); Project 5 holds 48 issue items, 0 pull requests. |
185
- | DF-27 — re-plan after apply proposes nothing | After `--apply` | #323 | observed 2026-10-26: fresh `--plan --out <scratch>` has 0 rows in batches 1 to 6. |
186
- | DF-28 — the record ships in the consolidated PR (D-93, D-112) | Record PR | #323 | observed 2026-10-26: PR #332 (`chore/323-projects-migration` into the integration branch) carries `workstream/migration-projects.md`; the consolidated PR references #332. |
187
- | DF-29 — PRD AC-12, FR-19, FR-20 on live data | After `--apply` | #323 | observed 2026-10-26: no D-91 retired label defined; #317 under #289 with #318 to #323; one marker comment on #289; titles unchanged. |
157
+ | Criterion | Event | Issue | Status |
158
+ | ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
159
+ | PRD AC-1 — publish creates the PRD/wave/story hierarchy | Wave 2 publish by `product-engineer` (DF-1, zero-match path) | #289, #290, #291 to #296 | observed 2026-10-21: wave 2 publish created the PRD and wave tracking issues and attached six stories, hierarchy verified(7 of 7) |
160
+ | PRD AC-2 (live half) — tracking issues carry `level:` only | Wave 2 publish (DF-1) | #289, #290 | observed 2026-10-21: both tracking issues carry their `level:` label only |
161
+ | PRD AC-3 (b) — story status comment, S-001 | `/developer` on #270 (DF-2) | #270 | n/a (pre-rule) |
162
+ | PRD AC-3 (b) — story status comment, S-002 | `/developer` on #271 (DF-2) | #271 | n/a (pre-rule) |
163
+ | PRD AC-3 (b) — story status comment, S-003 | `/developer` on #272 (DF-2) | #272 | observed 2026-10-20: one marked comment (id 6302983274), edited in place; comment created at sub-task 3.2a, after the draft PR, because the rule did not exist before 3.2 (S-003 DR-2) |
164
+ | PRD AC-3 (b) — story status comment, S-004 | `/developer` on #273 (DF-2) | #273 | observed 2026-10-20: one marked comment (id 6303270800), edited in place |
165
+ | PRD AC-3 (b) — story status comment, S-005 | `/developer` on #274 (DF-2) | #274 | observed 2026-10-20: one marked comment (id 6303586826), edited in place |
166
+ | PRD AC-3 (b) — story status comment, S-006 | `/developer` on #275 (DF-2) | #275 | observed 2026-10-21: one marked comment (id 6305768927), edited in place |
167
+ | PRD AC-3 (a, live half) — Project Status written only by automations | Consolidated PR to main opens with `Closes #N` (Status → In progress) and merges (→ Done); story PRs into the integration branch move nothing (D-50) (DF-5) | #270 to #275; #285, #290 to #296 | observed: wave 1 Done after PR #282 merged; wave 2 In progress for all eight issues at 20:39Z 2026-10-22 after consolidated PR #312 opened, read with `projectItems`, no agent write; Done for all eight at 21:00Z after PR #312 merged, `project_observation: observed` Wave 3: In progress at consolidated PR #335 open (observed 2026-10-27 12:50Z, #317 and #318) and Done at its merge (observed 2026-10-27 13:01Z, #317 to #323). |
168
+ | FR-24 — wave update, no-parent path (404) | Any wave 1 story's `/developer` run (DF-3) | #272, #273, #274, #275 | observed: `gh api …/parent` returns 404; closeout carries `wave_status: skipped(no parent)` |
169
+ | FR-24 — wave update, success path (200) | First main-thread `/developer` run on a wave 2 story: parent found, wave comment edited (DF-3) | pending | pending |
170
+ | D-36 — second publish reuses the PRD issue (one-match path) | Wave 3 publish (DF-1b) | #289, #317 | observed 2026-10-24: the wave 3 publish found exactly one open `[PRD-projects-tracking]` issue and reused #289, then created wave #317 under it (`workstream/github-publication-projects-tracking-phase-3.md`). |
171
+ | PRD AC-5 (live half) — runbook exercise (DF-4) | Owner follows `docs/runbooks/runbook-configure-github-project.md` on `llipe/dev-tasks` | #274 (Project 5) | observed 2026-10-21: Project 5 created from the Team planning template; seven shipped workflows enabled; auto-add `is:issue`; fields kept by decision D-48 |
172
+ | Wave 2 DF-6 — checkpoint committed after every merge | `/planner` wave 2 run (S-008 rule) | integration/projects-tracking-phase-2 | observed 2026-10-22, hand-applied (rule merged at S-008): six `chore(planner): checkpoint after S-0xx merge` commits, each after its merge |
173
+ | Wave 2 DF-7 — state file carries `Wave issue` and `PRD issue` | `/planner` wave 2 run (S-008, S-018 rules) | #290, #289 | observed 2026-10-22, hand-applied |
174
+ | Wave 2 DF-8 — one wave status comment, edited in place | `/planner` wave 2 run (S-009 rule) | #290 | partly observed 2026-10-22, hand-applied: one marker comment edited at run start and every merge; never edited at a story's PR-ready closeout, so no story showed `PR open` (rollup RD-1) |
175
+ | Wave 2 DF-9 — no per-merge table comments | `/planner` wave 2 run (S-009 rule) | #290 to #296 | observed 2026-10-22 |
176
+ | Wave 2 DF-11 — consolidated PR carries `Closes` for every merged story and the wave | Consolidated PR open (S-010 rule) | #312 | observed 2026-10-22: `Closes` for #291 to #296, #290 and #285 under `## Why`; GitHub links all eight |
177
+ | Wave 2 DF-15 — close-out runs from the merged prompt | Phase 6 in a fresh session on `main` (S-011 rule) | close-out PR | observed 2026-10-22: Phase 6 run from the merged prompt on `main` |
178
+ | Wave 2 DF-16 — PRD status comment at run start, PR open and close-out | `/planner` wave 2 run (S-018 rule) | #289 | run start hand-applied; PR open and shipped observed 2026-10-22 from the merged prompt |
179
+ | DF-21 — S-016 roadmap check is a no-op on a canonical roadmap | Wave 3 migration, before `--plan` | #323 | observed 2026-10-25: `roadmap-parity` 28/28; this repository's roadmap was already canonical, so no conversion PR. |
180
+ | DF-22 — `--plan` is read-only (FR-17) | Wave 3 migration, `--plan` | #323 | observed 2026-10-26: labels and open issues identical before and after a `--plan` run (qa-engineer, PR #332). |
181
+ | DF-23 — Migration Mode asks one question per review item (S-017/AC-2) | Wave 3 migration review walk | #323 | observed 2026-10-25 (planner attestation): nine items asked one per message; the owner overrode two recommendations (#108, #262). |
182
+ | DF-24 — `Approved: yes` set only on explicit approval (D-80) | Before `--apply` | #323 | observed 2026-10-25 (planner attestation): set after the owner wrote "approved"; one `Approved: yes` line; Plan digest recomputes to the header. |
183
+ | DF-25 — `--apply` records every write (S-015/AC-3, AC-8) | Wave 3 migration, `--apply` | #323 | observed 2026-10-25: exit 0, 14 done, 0 skipped, 0 failed; every row matches GitHub. |
184
+ | DF-26 — the Project batch is the owner's (PRD AC-12, D-81) | Owner runs batch 7 | #323 | observed 2026-10-26: four steps ticked by the owner (screenshot: auto-add `is:issue`, Item closed and Pull request merged on); Project 5 holds 48 issue items, 0 pull requests. |
185
+ | DF-27 — re-plan after apply proposes nothing | After `--apply` | #323 | observed 2026-10-26: fresh `--plan --out <scratch>` has 0 rows in batches 1 to 6. |
186
+ | DF-28 — the record ships in the consolidated PR (D-93, D-112) | Record PR | #323 | observed 2026-10-26: PR #332 (`chore/323-projects-migration` into the integration branch) carries `workstream/migration-projects.md`; the consolidated PR references #332. |
187
+ | DF-29 — PRD AC-12, FR-19, FR-20 on live data | After `--apply` | #323 | observed 2026-10-26: no D-91 retired label defined; #317 under #289 with #318 to #323; one marker comment on #289; titles unchanged. |
188
188
 
189
189
  ## Changelog
190
190
 
191
191
  | Version | Date | Change | Author |
192
192
  | ------- | ---------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------ |
193
+ | 1.13 | 2026-10-27 | DF-5: wave 3 In progress at consolidated PR #335 open and Done at its merge, both observed with no agent Project write. | planner |
193
194
  | 1.12 | 2026-10-27 | Dogfood checks: wave 3 rows DF-21 to DF-29 observed (#323, PR #332); the D-36 one-match publish row observed on the wave 3 publish (#289 reused). | planner |
194
195
  | 1.11 | 2026-10-24 | Fixtures and Mocking: the `node:fs` crash mock for EC-140 (S-015 task 5.22, #322). | developer |
195
196
  | 1.10 | 2026-10-24 | Fixtures and Mocking: the `--apply` harness (the model's write half, `apply-run.ts`, `random-repo.ts`, the fake `gh`'s `<file>` keys and delay) (S-015, #322). | developer |
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@llipe.com/dev-tasks",
3
- "version": "0.17.0",
3
+ "version": "0.17.1",
4
4
  "description": "AI agent toolkit for structured, PRD-driven development workflows",
5
5
  "type": "module",
6
6
  "imports": {