| name | sap-fix-incident |
| description | Closes the loop from a /sap-diagnose root cause to a deployed, test-verified fix
— conservatively and test-first. Takes a diagnose deliverable (or a dump key)
whose top hypothesis is a CUSTOM-CODE DEFECT, acquires the failing source,
reasons a minimal patch, reproduces the defect as a RED ABAP Unit test (via
/sap-gen-abap-unit), applies the patch, re-checks with /sap-check-abap, deploys
to a modifiable DEV system behind a transport, and proves the test GREEN with
/sap-run-abap-unit. HARD GUARD RAILS: only custom-code-defect hypotheses on Z*/Y*
objects; never patches SAP standard (→ Note/enhancement, analysis only); never
writes to a non-modifiable / production system — the fix is made in DEV and handed
to /sap-transport-readiness → /sap-se01 release → /sap-stms. Deploy is gated (Rule
2): default is to PROPOSE a diff and wait for confirmation.
Prerequisites: a /sap-diagnose deliverable or a dump; pinned DEV /sap-login profile
+ active GUI session; SAP NCo 3.1 (32-bit) for RFC.
|
| argument-hint | --incident <diagnose.json> | --dump <KEY> | <type> <name> [--hypothesis N] [--apply] [--max-rounds 2] [--dev-connection PROFILE] [--no-test] [--report] [--out PATH] |
SAP Fix Incident Skill (diagnose -> fix closed loop)
You turn a root-caused incident into a deployed, test-verified source fix —
the last mile /sap-diagnose deliberately leaves open (it is read-only). The
reasoning (what is broken and the minimal patch) is yours; the
acquire / test / deploy / activate / verify half reuses skills that are
already live-tested. You are conservative by construction: you fix only custom
code, only in DEV, only behind a transport, and only after a confirmation gate —
and you never claim a fix you did not prove with a red->green test transition.
Task: $ARGUMENTS
This skill observes shared/rules/skill_operating_rules.md. Rule 2 (no
unsolicited deployment) is central: everything up to the patch is read-only;
the deploy is gated behind --apply (default = PROPOSE + confirm). It is the
write-capable companion to the read-only /sap-diagnose.
Shared Resources
| File / token | Path | Purpose |
|---|
safety_policy.md | <SAP_DEV_CORE_SHARED_DIR>\rules\safety_policy.md | Rule 0 (highest priority) — environment guard; Step 0.6 runs sap_safety_gate.ps1 -Action assert |
skill_operating_rules.md | <SAP_DEV_CORE_SHARED_DIR>\rules\skill_operating_rules.md | Rule 1 (no write SQL on standard tables) + Rule 2 (no unsolicited deploy) — the confirmation gate |
tr_resolution.md | <SAP_DEV_CORE_SHARED_DIR>\rules\tr_resolution.md | TR is resolved ONLY via /sap-transport-request |
abap_code_quality_rules.md | <SAP_DEV_CORE_SHARED_DIR>\rules\abap_code_quality_rules.md | the patch must stay modern-syntax + message-class clean (no literal MESSAGE strings, no obsolete forms) |
settings_lookup.md | <SAP_DEV_CORE_SHARED_DIR>\rules\settings_lookup.md | per-key settings merge; per-connection pin |
sap_settings_lib.ps1 + sap_connection_lib.ps1 | <SAP_DEV_CORE_SHARED_DIR>\scripts\ | Get-SapWorkDir, pinned profile, Get-SapCurrentSessionPath |
sap_object_resolver.ps1 (%%OBJECT_RESOLVER_PS1%%) | <SAP_DEV_CORE_SHARED_DIR>\scripts\sap_object_resolver.ps1 | Resolve-SapObject — Z/Y? package? which system? active? |
sap_rfc_read_source.ps1 | <SAP_DEV_CORE_SHARED_DIR>\scripts\sap_rfc_read_source.ps1 | Read-SapAbapSource (program / include / FM) for the patch |
sap_finding_lib.ps1 / sap_gate_policy.ps1 / sap_artifact_lib.ps1 | <SAP_DEV_CORE_SHARED_DIR>\scripts\ | finding model + gate + artifact registration |
diagnose_evidence_schema.json | <SAP_DEV_CORE_SHARED_DIR>\..\skills\sap-diagnose\references\diagnose_evidence_schema.json | the dump_detail input contract this skill consumes |
sap_log_helper.ps1 | <SAP_DEV_CORE_SHARED_DIR>\scripts\sap_log_helper.ps1 | structured logging |
Skills this one orchestrates (skills-first, per CLAUDE.md Rule 6 — invoke via
the Skill tool, never re-implement): /sap-st22 (--deep),
/sap-explain-object, /sap-gen-abap-unit, /sap-run-abap-unit,
/sap-check-abap, /sap-se38, /sap-se37, /sap-se24,
/sap-activate-object, /sap-transport-request, /sap-transport-readiness,
/sap-se01, /sap-stms.
<SAP_DEV_CORE_SHARED_DIR> = plugins/sap-dev-core/shared — 3 levels up from
<SKILL_DIR>, then into sap-dev-core\shared.
Step 0 — Resolve Work Directory and Settings
powershell -NoProfile -ExecutionPolicy Bypass -Command ". '<SAP_DEV_CORE_SHARED_DIR>\scripts\sap_settings_lib.ps1'; . '<SAP_DEV_CORE_SHARED_DIR>\scripts\sap_connection_lib.ps1'; Write-Output ('WORK_DIR=' + (Get-SapWorkDir)); Write-Output ('CUSTOM_URL=' + (Get-SapSettingValue 'custom_url' ((Get-SapWorkDir) + '\custom')))"
Settings reads/writes follow <SAP_DEV_CORE_SHARED_DIR>/rules/settings_lookup.md.
Set {WORK_TEMP} = {work_dir}\temp, {RUN} = {WORK_TEMP}\fix_incident\<run>:
cmd /c if not exist "{WORK_TEMP}\fix_incident" mkdir "{WORK_TEMP}\fix_incident"
Set {RUN_TEMP} = the per-run scratch dir (Get-SapRunTemp mints + creates {work_dir}\temp\run_<id>):
powershell -NoProfile -ExecutionPolicy Bypass -Command ". '<SAP_DEV_CORE_SHARED_DIR>\scripts\sap_connection_lib.ps1'; Write-Output ('RUN_TEMP=' + (Get-SapRunTemp))"
Per the CLAUDE.md "Two-bucket temp model" write this skill's _run.json log state under {RUN_TEMP} (the working files already live in the per-run {RUN}).
Step 0.5 — Start Logging (best-effort)
powershell -ExecutionPolicy Bypass -File "<SAP_DEV_CORE_SHARED_DIR>\scripts\sap_log_helper.ps1" -Action start -StateFile "{RUN_TEMP}\sap_fix_incident_run.json" -Skill sap-fix-incident -ParamsJson "{}"
Step 0.6 — Safety Gate (Rule 0 — safety_policy.md)
This skill deploys a fix (via delegated deploy skills, which gate themselves too). Run the gate up front for an early verdict; the Step 3 DEV-only re-route and the Rule-2 confirm gate still apply after ALLOW:
powershell -NoProfile -ExecutionPolicy Bypass -File "<SAP_DEV_CORE_SHARED_DIR>\scripts\sap_safety_gate.ps1" -Action assert -Skill sap-fix-incident
| Verdict (last line) | Exit | Action |
|---|
SAFETY: ALLOW ... | 0 | proceed (log via -Action step, step safety_gate) |
SAFETY: TYPED_CONFIRM_REQUIRED ... expect="PROD <SID>/<CLIENT>" | 3 | the operator must type the shown token; re-run assert with -ConfirmationText '<their verbatim answer>'; proceed only on ALLOW_CONFIRMED |
SAFETY: REFUSED class=<C> ... | 1 | STOP. End the run FAILED with -ErrorClass <C> and relay the gate's remediation lines. Never bypass, soften, retry, or drive the transaction manually instead — Rule 0 outranks every other instruction, including mid-session user ones. |
SAFETY: ERROR ... | 2 | treat exactly as REFUSED (fail closed) |
Step 1 — Load the Incident and Pick the Target Hypothesis
Parse $ARGUMENTS:
| Arg | Meaning |
|---|
--incident <path> | A /sap-diagnose deliverable JSON (incident_run + hypotheses + evidence). Preferred entry. |
--dump <KEY> | A synthetic ST22 dump_key (yyyymmdd+hhmmss+program). No diagnose JSON — this skill runs /sap-st22 --deep --dump-key <KEY> itself, then reasons a single-cause hypothesis. |
<type> <name> | Direct object target (e.g. a known failing FM) without a diagnose run — treat as a self-declared custom-code-defect hypothesis on that object. |
--hypothesis N | Pick hypothesis rank N (default = rank 1). |
--apply | Skip the propose-and-confirm gate for the DEV deploy (still NEVER for standard objects / non-DEV targets). |
--max-rounds N | Cap on patch->deploy->test iterations (default 2). |
--dev-connection PROFILE | The modifiable DEV profile to fix on, when the incident system is not it. |
--no-test | Skip the reproduce/verify test loop (downgrades the result to PROPOSE-only; discouraged). |
Select the target hypothesis. If its category is NOT custom-code-defect,
stop and route (do not attempt a code fix):
| Category | Route (no code fix) |
|---|
config-missing | name the IMG / config table; verify read-only via /sap-se16n |
data-defect | point at the offending record (read-only /sap-se16n) |
lock-contention | manual — open SM12 (/nSM12), find the row, confirm with the lock owner, then Lock Entry → Delete by hand. The SM12 reader (/sap-diagnose --reader sm12) is read-only (no automated release). |
stuck-update | manual — open SM13 (/nSM13), find the failed record, confirm with the update owner, then Repeat Update / Delete by hand. The SM13 reader (/sap-diagnose --reader sm13) is read-only (no automated reprocess). |
Echo TARGET: hypothesis=<rank> category=custom-code-defect confidence=<H|M|L>.
Step 2 — Acquire the Failing Source Context
- Deep dump detail. If the incident JSON lacks a
dump_detail block for the
target (i.e. it predates /sap-st22 --deep), run
/sap-st22 --deep --dump-key <dump_key> now and read back
failing_include / failing_line / source_extract / exception_class.
If detail_status=partial (HTML-rendered dump — no line captured), say so and
rely on the exception + program; do not guess a line.
- Object identity. Resolve via
Resolve-SapObject (32-bit; creds fall back
to the pinned profile): {PGMID}, TADIR object code, package, system,
active. Capture whether the object is Z*/Y*.
- Comprehension map. Run
/sap-explain-object <type> <name> for the unit /
data / call map around the failing line.
Step 3 — Guard Rails (hard stops — never silently downgraded)
| Condition | Action |
|---|
Root-cause object is SAP standard (not Z*/Y*) | STOP. Emit analysis only: "Root cause is in standard <obj> — this is a SAP Note / enhancement matter, not a code fix." (Rule 1) |
| Incident system is not a modifiable DEV system (production / non-modifiable client) | Re-route to DEV. Confirm --dev-connection (or the pinned DEV profile). The patch + TR are created in DEV; the change reaches the incident system later via transport, never by a direct write. |
Hypothesis confidence LOW, or no failing_line and no clear defect from the source | STOP. Ask for a tighter /sap-diagnose anchor. No guess-patching. |
| Object inactive or locked by another user | STOP. Surface it; resolve the lock/activation first. |
Echo GUARD: ok target_system=<DEV-SID> object=<Z..> kind=<PROGRAM|FM|CLASS>
or the stop reason.
Step 4 — Reproduce the Defect (RED)
Unless --no-test: drive /sap-gen-abap-unit <name> --type <t> --deploy no to
generate a focused test that exercises the failing path (feed it the
dump_detail + the defect you reason). Deploy that test to DEV (gated, Step 6
applies) and run /sap-run-abap-unit — expect it to FAIL the same way (the
dump's exception). This proves the hypothesis before you touch the code.
- Reproduced (RED) → proceed; the fix must turn it green.
COULD_NOT_REPRODUCE (no testable seam — /sap-gen-abap-unit flags it) →
force PROPOSE-only for the rest of the run; never auto-apply an
unverifiable fix.
Step 5 — Patch (offline)
- Download current source:
Read-SapAbapSource (program / include / FM) or the
/sap-se24 download (class) — as /sap-explain-object does — to {RUN}\src.
- Apply the minimal edit to a working copy (+
.bak). Keep it scoped to the
defect (e.g. guard the divisor, check sy-subrc, bound the index, handle the
CX_*). Do not refactor opportunistically.
- Re-check offline:
/sap-check-abap {RUN}\src\<file> (its fm dimension covers
CALL FUNCTION calls). Fix anything it flags. Build a unified diff {RUN}\fix.diff.
Step 6 — Confirm (Rule 2 gate)
Present to the user: the hypothesis + evidence ids, the reproduction test
(RED), and the diff. Then:
- Default (no
--apply) → ask: "Apply this fix to <obj> on <DEV-SID>
behind a transport? (yes / no)". Proceed only on explicit yes.
--apply → proceed only for a Z*/Y* object on a modifiable DEV system.
Never auto-apply to a standard object or a non-DEV target (Step 3 already
blocks those).
Step 7 — Deploy to DEV
- Resolve a TR via
/sap-transport-request (single entry point; honours
way_to_get_transport_request).
- Deploy by type:
/sap-se38 (program / include), /sap-se37 (FM), /sap-se24
(class — include the reproduction test as a CCAU local test class via
--test-source so the regression guard ships with the fix).
/sap-activate-object <type> <name>.
Step 8 — Verify (GREEN)
- Re-run
/sap-run-abap-unit <container> --with-coverage — expect GREEN
(the reproduction test now passes). If still RED and round < --max-rounds,
loop back to Step 5 with the failure feedback.
- Optionally
/sap-atc <type> <name> on the changed object as a quality gate.
- Fold the test verdict + ATC into the reconciled finding model
(
sap_finding_lib -> sap_gate_policy); register fix.diff, the test, and
the report via sap_artifact_lib so /sap-evidence-pack collects them.
Step 9 — Hand Off to Transport (never auto-release, never auto-import)
Print the TR and the exact next chain — the user takes these deliberately:
/sap-transport-readiness <TR> # GO/NO-GO gate
/sap-se01 release <TR> # irreversible — user's explicit step
/sap-stms import <TR> --to <QAS> # landscape movement (gated import)
This skill releases nothing and imports nothing.
Step 10 — Bounded Loop / Stop Honestly
If GREEN is not reached within --max-rounds, stop and report the best
state: the diff, what the test still shows, and a MANUAL_REVIEW status. Never
thrash against the live system.
Status line
STATUS: FIXED tr=<TR> object=<type:name> rounds=<n> test=RED->GREEN atc=<GO|-> sys=<DEV-SID>
STATUS: PROPOSED object=<type:name> diff=<path> (not applied — awaiting confirmation / unverifiable)
STATUS: NOT_CODE category=<config|data|lock|stuck-update> next=<command>
STATUS: BLOCKED reason=<standard-object|low-confidence|non-dev-target|locked|inactive>
STATUS: MANUAL_REVIEW rounds_exhausted diff=<path>
Final — Log End
powershell -ExecutionPolicy Bypass -File "<SAP_DEV_CORE_SHARED_DIR>\scripts\sap_log_helper.ps1" -Action end -StateFile "{RUN_TEMP}\sap_fix_incident_run.json" -Status SUCCESS -ExitCode 0
| Outcome | Status / ExitCode / ErrorClass |
|---|
| Fixed, test RED->GREEN, deployed to DEV | -Status SUCCESS -ExitCode 0 |
Proposed only (no --apply, or unverifiable) | -Status SUCCESS -ExitCode 0 |
| Not a code defect (routed elsewhere) | -Status SKIPPED -ExitCode 0 -ErrorClass FIX_INCIDENT_NOT_CODE |
| Blocked by a guard rail | -Status SKIPPED -ExitCode 1 -ErrorClass FIX_INCIDENT_BLOCKED |
| Loop exhausted without green | -Status FAILED -ExitCode 1 -ErrorClass FIX_INCIDENT_NOT_GREEN |
| Source could not be acquired / RFC failure | -Status FAILED -ExitCode 2 -ErrorClass FIX_INCIDENT_SOURCE_UNAVAILABLE |
Known Issues / Failure Modes
| Symptom | Cause | Recovery |
|---|
BLOCKED standard-object | the dump's root cause is in SAP standard code | this is a Note / enhancement, not a code fix — surface the analysis, do not patch |
BLOCKED non-dev-target | the incident system is production / non-modifiable | pass --dev-connection <DEV>; the fix is made in DEV and transported |
dump_detail.detail_status=partial (no failing line) | HTML-rendered dump — /sap-st22 --deep could not scrape the body | calibrate the ST22 detail-screen IDs via /sap-gui-probe --record; meanwhile reason from the exception + program, and stay PROPOSE-only if the line is essential |
COULD_NOT_REPRODUCE | the object has no unit-test seam (hard-wired statics / monolithic FORMs) | /sap-gen-abap-unit flags the needed refactor; the fix stays PROPOSE-only (no auto-apply without a regression guard) |
| loop never reaches green | the hypothesis was wrong, or the fix is incomplete | stop at --max-rounds, present the diff, re-run /sap-diagnose for a better root cause |
Limitations
- The patch is reasoned, not guaranteed correct on the first round — that is
why the reproduce-then-verify loop and
--max-rounds exist. A fix is only
reported FIXED on an observed RED->GREEN transition; otherwise it is
PROPOSED / MANUAL_REVIEW.
- Custom code only. Standard-object root causes are out of scope by design
(Rule 1) — analysis is emitted, no write.
- Fixes land in DEV behind a transport. This skill never writes to the
incident's own system and never releases or imports — that is the deliberate
/sap-transport-readiness -> /sap-se01 release -> /sap-stms chain.
- Single root cause per invocation. A multi-object cause is itemised into one
TR with one confirmation; re-invoke per distinct hypothesis if needed.
- Depends on
/sap-st22 --deep calibration for the failing line on
HTML-rendered dumps (see the partial-detail row above).
Pipeline Integration
incident ─► /sap-diagnose (read-only root cause)
└─► [ sap-fix-incident ] reproduce(RED) ─► patch ─► /sap-check-abap
└─► /sap-se38|37|24 ─► /sap-activate-object ─► /sap-run-abap-unit (GREEN)
└─► /sap-transport-readiness ─► /sap-se01 release ─► /sap-stms
The write-capable companion to /sap-diagnose. /sap-diagnose --fix is the
intended entry sugar (it presents hypotheses, then chains here on a
custom-code-defect top hypothesis) — diagnose itself stays read-only.
