Index of every page on this site: llms.txt
All skills
akdevsaha-dev avatar

/catch-up

@ce26f0a
Run it yourself

Explain what teammates changed in a shared Git repository since you last worked in it. Finds your last commit as the baseline, reads the real diffs and surrounding code, groups changes by feature, and flags what affects your current work, breaking changes, migrations, risks, and what to review. Read-only. Use when the user asks to catch up, what changed, what happened while they were away, what teammates did, what landed on main, or what a pull would bring in.

Use this Skill: https://skilld.dev/gh/akdevsaha-dev/factory-skills/catch-up

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈119 tokens for metadata: the name and description. ≈2.8k when used: this file. ≈4.7k more on demand in 2 files.

Description uses 5.9% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈475 characters, or 5.9%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Catch Up

Explain what happened in the repository since the user last looked, the way a senior teammate would brief them: grouped by feature, grounded in the actual code, and focused on what affects their work and what deserves a closer look.

Ground rules

  • Read-only. Never modify the working tree, the index, refs, or history. No fetch, pull, merge, rebase, checkout, switch, stash, reset, commit, or file edits, not even temporarily. references/git-history.md lists the safe commands. The only write is save, and only when the user asks for it.
  • Never fetch on your own. Remote analysis uses the remote-tracking refs as they are. Fetch only when the user explicitly asks for it in this conversation.
  • The diff is the evidence. Commit messages, PR titles, and branch names are claims to check against the code, never the source of a finding.
  • Count each change once. Describe the net change of the range. Use individual commits only for attribution, grouping, and intent.
  • Separate fact from inference. Label what the code shows as confirmed and what you deduce about intent or impact as inferred.
  • Cite locations. Point to path:line in the target version for every specific claim.
  • Signal over completeness. Group related work, collapse trivial changes into one line, and skip what does not matter.

Options

Parse these from the arguments or plain words ("since Monday", "only Bob's changes"). They combine; for example, /catch-up deep remote author bob.

Option Effect
quick (default) High-signal changes read in full, everything else skimmed
deep Every meaningful diff read with surrounding code, callers, tests, and merge resolutions
remote [<ref>] Analyze the remote-tracking branch (the upstream by default) instead of local HEAD
since <commit, ref, or date> Baseline is that commit, or the last commit before that date
last <N> Baseline is N changes back on the target's first-parent line
checkpoint Baseline is the checkpoint saved for this branch
author <name or email> Only changes authored by that person
save After reporting, save the analyzed commit as this branch's checkpoint

Step 1: Resolve the range

Read references/git-history.md, then determine:

  1. Your work. Current branch, HEAD, uncommitted changes, and local commits not on the upstream. Read uncommitted changes only with the commands in "Your uncommitted work" in the reference; a plain git diff rewrites the index.
  2. Target. Local HEAD, or with remote, the remote-tracking ref, its last fetch time, and how far local is ahead and behind. In local mode, still note in one line when the upstream has commits that local lacks.
  3. Baseline. An explicit since, last, or checkpoint wins. Otherwise use your most recent authored commit reachable from the target, matched by your Git identity. Unpushed commits are not reachable from a remote target, so in remote mode the baseline is your last commit that the remote has.
  4. Range. All commits from the baseline to the target, with the author filter applied.

Show the resolved range before the analysis, and continue without waiting:

Catch-up  main · local HEAD 8e41a2c · quick
Baseline  51c09fd "Add order export", your last commit (4 days ago)
          Override with: since <ref or date>, last <N>, checkpoint, author <who>
Range     14 commits in 5 changes by Priya Shah, Tom Okafor · 23 files, +612 −148

When the range is empty, show the header with Range empty and stop. Then say whether the remote-tracking branch has commits that local lacks, as of its last fetch (suggest /catch-up remote). When no baseline can be found, say why and offer since or last with a suggested value.

Step 2: Map the history

Follow "Count each change once" in references/git-history.md:

  1. List the units of change: the target's first-parent commits in the range. Each is a direct commit, a squash merge, or a merge commit whose branch is one unit.
  2. Take the net diff from baseline to target, with rename detection, as the single account of what is now different.
  3. Note reverts and changes added and reverted inside the range, duplicates from cherry-picks or rebases, and merges with manual resolutions.

Step 3: Read the changes

Read references/review-guide.md. Classify every changed file as high-signal, ordinary, or generated. Then read by mode:

quick deep
High-signal files (migrations, contracts, routes, exported interfaces, auth, config, dependencies, CI, deletions) Full diff, plus the surrounding code at the target Same
Ordinary files Stat and hunk headers; open a hunk when its context suggests it matters Full diff with surrounding code
Consumers of changed interfaces git grep at the target for callers of changed or removed symbols Trace callers and callees through the code
Merge commits Resolution check on merges that touch high-signal files Resolution check on every merge
Tests Whether behavior changes came with tests Read the tests and judge what they cover
Generated files and lockfiles Lockfiles only, to confirm versions Same

Read code at the target with git show <target>:<path>. Never check out. In deep mode on a large range, hand each feature group to a read-only subagent if your environment has one, and keep only its conclusions.

When the range is large (hundreds of commits or thousands of changed lines), keep quick to the high-signal files, summarize the rest by area, and suggest narrowing with since, last, or author.

Step 4: Group by feature

Cluster the units and files into feature groups: one user-facing capability or one technical concern each (for example "rate limiting on user lookups", not "Bob's commits"). A unit that does two unrelated things belongs to two groups, and several units that build one thing form one group.

For each group, record what changed (confirmed), why it was likely done (inferred), who did it, the units involved, and the key locations. Cover the change types in references/review-guide.md: architecture, API contracts, migrations, dependencies, refactors, and removed functionality.

Step 5: Assess impact on your work

Follow "Impact on your work" in references/review-guide.md. Check:

  • Textual conflicts. In remote mode, your unpushed commits against the remote, using the isolated merge check in references/git-history.md. Your uncommitted files against the range's files, by comparing hunks.
  • Semantic breaks. Your code calling an interface that changed, a file you edit that moved, migrations that collide with yours, new required configuration, and dependencies you need to install.

In local mode the range is already in your HEAD, so the question is whether your uncommitted work still fits it. In remote mode the question is what a pull or rebase would bring and break.

With author, assess only that person's changes, but if something outside the filter breaks your work, add one line saying so.

Step 6: Find risks

Look for potential bugs, security concerns, breaking changes, and missing verification, using the checklists in references/review-guide.md. Report a risk only with a concrete location and a scenario in which it goes wrong. Rate each risk High, Medium, or Low by its likely impact. When it depends on something you cannot see, such as the production schema or an entry point outside the repository, say so in the finding.

Step 7: Report

Lead with a short summary, then important changes, then impact, then what to review. Keep it scannable. Do not narrate commits one by one.

Catch-up  main · local HEAD 8e41a2c · quick
Baseline  51c09fd "Add order export", your last commit (4 days ago)
Range     14 commits in 5 changes by Priya Shah, Tom Okafor · 23 files, +612 −148

Summary
  Checkout now creates a payment intent before saving the order, and order totals are stored in
  cents, with a migration that rewrites existing rows. The server now requires STRIPE_WEBHOOK_SECRET.
  Your uncommitted export code still reads the removed order.total field.

Important changes
  1. Payments move to intents (Priya Shah, merge 8e41a2c, #88)
     Confirmed  createOrder() calls payments.createIntent() before saving (src/orders/service.ts:42)
     Confirmed  New POST /webhooks/stripe verifies the signature (src/webhooks/stripe.ts:12)
     Confirmed  src/config.ts:18 throws when STRIPE_WEBHOOK_SECRET is unset; src/server.ts:3 imports it
     Inferred   Payment confirmation becomes asynchronous, so orders can stay pending
  2. Order totals in cents (Tom Okafor, squash 3fa90b1, #91)
     Confirmed  Order.total (decimal) replaced by totalCents (integer) (src/orders/model.ts:9)
     Confirmed  Migration rewrites every row in place, with no down migration (db/migrations/0042_total_cents.sql:3)
  3. Smaller: logger moved to src/lib/log.ts; eslint upgraded to v9 (config rewritten, same rules)

Impact on your work
  ✗ src/orders/export.ts:27 (uncommitted) reads order.total, which no longer exists (src/orders/model.ts:9)
  ✗ db/migrations/0042_export_log.sql (untracked) has the same number as 0042_total_cents.sql
  ! Set STRIPE_WEBHOOK_SECRET locally or the server will not start
  ✓ No textual overlap with your other uncommitted changes

Review
  1. High    src/webhooks/stripe.ts:30  marks the order paid without comparing the event amount to the order
  2. Medium  db/migrations/0042_total_cents.sql:3  one UPDATE over all orders holds a table lock
  3. Low     src/orders/service.ts:42  the new pending state has no tests

Not checked  CI results; whether production sets STRIPE_WEBHOOK_SECRET

In remote mode, the header gains the local and remote state, Range covers the whole range, and Split divides it. Mark each change · already local or · remote-only after its title:

Catch-up  main · remote origin/main 8e41a2c · deep
Local     main @ 22a29fb · 1 ahead, 4 behind origin/main · 2 uncommitted files
Remote    origin/main @ 8e41a2c · last fetched 2 hours ago (not fetched now)
Split     1 change already local · 4 remote-only

With author, Range shows that person's commits and their diff stats, followed by the whole range's counts in parentheses. With last <N>, the Baseline line reads <sha>, last N changes (a merged branch counts as one).

Omit empty sections. If you saved a checkpoint, end with the ref you wrote.

Checkpoints

A checkpoint records where you last caught up, so the next run can start there. Save one only when the user asks (save), after the report, using the commands in "Checkpoints" in references/git-history.md. Tell the user the ref you wrote and how to delete it.

The checkpoint marks the whole range as seen. After an author-filtered run, other people's changes were not reviewed, so say that and ask before saving. If the user still wants it, record the filter in the checkpoint's message, and repeat that warning whenever the checkpoint is used.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts8d3 checks · Risk SAFE
  • Gen Agent Trust Hub8d

    The skill analyzes Git repository changes and provides a summary. It is generally safe but processes external code (Git diffs) which could contain malicious instructions (Indirect Prompt Injection). It also uses shell commands for Git operations.

  • Socket8d

    No alerts

  • Snyk8d

    Risk: LOW · No issues

Signed by skilld at ce26f0a. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated last week

README badge

README badge for akdevsaha-dev/factory-skills/catch-up